Penetration testing has always been a deeply human craft. Skilled security professionals spend hours, sometimes days, manually probing networks, enumerating services, and chaining together obscure software flaws into working exploits. As information systems grow increasingly complex and cyberattack techniques continue to evolve, this reliance on manual expertise faces serious challenges in both efficiency and scalability. The shortage of qualified pentesters means most organizations simply cannot test their systems as often or as thoroughly as the threat landscape demands. PentestMCP was designed to close that gap by letting language models orchestrate the entire offensive security workflow, from the first reconnaissance scan to the final exploitation payload.

The technical heart of the framework rests on three synergistically coupled mechanisms. The first is MCP-centered pentesting tool orchestration. The Model Context Protocol, or MCP, is an emerging standard for connecting language models to external software tools in a structured, standardized way. Rather than forcing an AI to invent shell commands from scratch and hope they work, PentestMCP wraps established security tools, the kind found in the Kali Linux distribution, behind standardized capability invocations. The system can call network scanners such as Nmap, SQL injection tools such as Sqlmap, and exploitation frameworks such as Metasploit through a consistent interface, which dramatically reduces the hallucination and formatting errors that plague naive LLM tool use.

The second mechanism is dynamic closed-loop scheduling built on what the authors call a PTG, a pentesting task graph that the agents continuously decompose, update, and reflect upon. Instead of following a rigid, pre-programmed checklist, the system treats each engagement as a directed graph of tasks that evolves as new information arrives. When a reconnaissance agent discovers an unexpected service or an unusual software version, the graph is rewritten on the fly, new subtasks are spawned, and completed branches are pruned. This closed loop of task decomposition and reflection allows the agents to behave less like scripts and more like human operators who adapt their strategy mid-engagement, a property the authors demonstrate in detailed appendices showing how the task graph for a well-known Hack The Box challenge called Shocker was initialized, updated, and finalized across the reconnaissance, enumeration, and exploitation stages.

The third and perhaps most novel mechanism is a dual-path exploitation policy featuring continuous learning. When the system encounters a vulnerability, it dynamically balances two strategies: precise knowledge retrieval from an existing exploit database and generative adaptability, where the language model reasons from first principles about how a flaw might be abused. If the retrieval path finds a known exploit pattern, the system executes it with high precision. If the vulnerability is unfamiliar, the generative path attempts to construct a novel exploitation strategy. Crucially, whenever the generative path succeeds in producing a new exploit path, that success is deposited back into the knowledge base, meaning the system literally teaches itself with every engagement. Over time, the fraction of vulnerabilities that require creative reasoning shrinks as the library of validated exploits grows.

The evaluation was extensive and grounded in reality rather than synthetic benchmarks. The team tested PentestMCP against more than one hundred real-world vulnerabilities spanning diverse CWE categories, the Common Weakness Enumeration taxonomy that classifies software flaws by type, and varying complexity levels. Experimental results show that PentestMCP consistently achieves higher success rates, execution accuracy, and efficiency than existing baselines while also reducing execution time. The baselines included earlier LLM-driven pentesting systems such as PentestGPT, which was evaluated at the USENIX Security Symposium, as well as multi-agent collaborative frameworks like Vulnbot and PentestAgent. The consistent outperformance across all three pentesting phases, reconnaissance, vulnerability discovery, and exploitation, suggests the architectural choices, standardized tool invocation, adaptive scheduling, and dual-path learning, compound rather than merely add up.

The phase-by-phase numbers deserve scrutiny because they reveal where AI currently excels and where it still struggles. Information gathering, at 87.2 percent success, is the phase where language models shine brightest, since it involves calling scanners, parsing outputs, and synthesizing findings, tasks well suited to LLM reasoning over structured tool responses. Vulnerability discovery, at 74.0 percent, requires correlating observed software versions and configurations with known CVE entries, a knowledge-intensive task where retrieval-augmented generation helps but hallucination risks remain. Exploitation, at 64.9 percent, is the hardest phase, since it demands chaining multiple steps, crafting payloads, and adapting to defensive measures in real time. The gradient across phases mirrors the expertise gradient of human pentesters, which is precisely why the researchers consider the results a meaningful benchmark of genuine capability rather than a parlor trick.

Underneath the headline results lies a sophisticated engineering stack. The knowledge base is backed by MongoDB, exploits and tool outputs are embedded using the all-MiniLM-L6-v2 sentence-transformer model and matched via cosine similarity, and task graphs are ordered using topological sorting over directed acyclic graphs. The dual-path policy draws on retrieval-augmented generation, the technique popularized by Lewis and colleagues at Facebook AI Research that grounds LLM outputs in retrieved documents rather than relying purely on parametric memory. By combining RAG with generative exploration, PentestMCP hedges against the two failure modes that have historically limited AI in security work: rigid systems that cannot handle novel situations and flexible systems that hallucinate plausible but wrong commands.

Yet the same capabilities that make PentestMCP valuable to defenders make it a potential weapon in the wrong hands. The researchers are candid about the dual-use dilemma, citing the growing literature on the malicious use of artificial intelligence and the taxonomy of risks posed by language models. Prompt injection attacks, where adversaries embed malicious instructions in data that LLM agents process, represent a particularly acute threat to agentic security systems, and recent work on indirect prompt injection in the wild and attacks on tool selection in LLM agents shows these risks are far from theoretical. There is also the alignment concern: as self-evolving LLM agents accumulate capabilities through continuous learning, ensuring they remain within authorized boundaries becomes an open research problem. The OWASP Top 10 for Large Language Model Applications now explicitly catalogs these threats, and the PentestMCP authors acknowledge that hallucination in large language models remains a fundamental challenge for any system that grants AI agents real offensive tooling.

What emerges from this study is a vivid picture of where cybersecurity is heading. Automated penetration testing driven by language models is no longer speculative; it is measurable, reproducible, and, in the case of PentestMCP, already outperforming earlier baselines across every phase of the offensive workflow. For defenders, systems like this promise to democratize rigorous security testing, letting organizations with modest budgets probe their infrastructure as often as needed rather than once a year. For attackers, the same frameworks lower the barrier to sophisticated intrusion. The research team, funded by the National Natural Science Foundation of China and several provincial programs, frames their work as a validation of substantial potential for advancement, but the deeper message may be more unsettling: the arms race between AI-powered offense and AI-powered defense has begun in earnest, and the systems that teach themselves with every hack are only going to get better.

Subject of Research: PentestMCP: LLM and MCP based multi-agent framework for automated penetration testing

Article Title: PentestMCP: LLM and MCP based multi-agent framework for automated penetration testing

Article References: Zhai, J., Zhou, X., Miao, H., Li, Z., Li, Z., & Yang, H. (2026). PentestMCP: LLM and MCP based multi-agent framework for automated penetration testing. Cluster Computing, 29(15), Article 854. https://doi.org/10.1007/s10586-026-06700-x

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06700-x

Keywords: cybersecurity, artificial intelligence, penetration testing, large language models, Model Context Protocol, multi-agent systems, vulnerability discovery, PentestMCP, based, multi-agent, framework, automated

The core idea is deceptively simple: rather than building a single monolithic program to mine the literature, the researchers assembled a set of LLM-based agents, each equipped with tools for specific tasks. One set of tools fetches omics studies and extracts article metadata. Another identifies and downloads published data files, whether raw instrument output or processed abundance tables. A third executes containerized quantification pipelines, so that the agents can re-derive abundance estimates from raw data rather than trusting deposited summaries. Finally, synthesis tools allow the agents to compare results across studies, including formal statistical combination of findings. The agents coordinate these capabilities through the Model Context Protocol, or MCP, a standard way of exposing containerized software as callable services so that a language model can invoke complex bioinformatics workflows without the workflow logic being hard-coded into the model itself.

Containerization is a crucial design choice. Reproducibility crises in computational biology often trace back to environment drift: a pipeline that ran on one machine with one version of a proteomics search engine may produce subtly different results elsewhere. By wrapping each analysis tool in a container with pinned dependencies, the framework ensures that whatever the agent decides to run, it runs the same way every time. The agent’s contribution is then decision-making—choosing which study to examine, which files to retrieve, which pipeline is appropriate for a given data type, and how to interpret the output—while the numerical heavy lifting remains the province of established, auditable software.

To test the system at scale, the researchers first ran the pipeline across thousands of articles in PubMed Central, cataloguing references to datasets and recording where published data could actually be located. The authors are careful about what this corpus-scale exercise demonstrates. They present the catalogue as a descriptive system output rather than a validated measurement of extraction accuracy, an honest framing that distinguishes feasibility from benchmarked performance. Even so, the exercise illustrates the kind of literature-wide survey that would be impractical to perform manually: a systematic map of which papers point to which repositories, and how often the essential files are actually findable.

The more demanding test came in the form of five end-to-end reanalyses, spanning data-dependent acquisition proteomics, data-independent acquisition proteomics, and bulk RNA sequencing. In each case, the agents had to locate the relevant publication, identify the deposited raw data, download it, execute an appropriate quantification pipeline, and compare their re-derived abundances against the values the original authors had deposited. All five reanalyses completed successfully, though the authors document that each required some human guidance and workflow accommodations along the way. This is an important caveat: the system is agent-supported rather than fully autonomous, and the paper does not claim otherwise. The human role was documented rather than hidden, which makes the results interpretable as a realistic picture of what current agentic systems can and cannot do unaided.

The quantitative agreement between agent-run pipelines and the original analyses is striking. Per-sample correlations between re-quantified and deposited abundances ranged from 0.85 to 0.997, indicating that the agents’ pipelines recovered essentially the same biological measurements. For differential expression—the comparison of features between experimental conditions that is usually the point of such studies—the fold-change concordance between reanalysis and original was measured by Spearman correlation at 0.88 to 0.91. Among features called differentially expressed in both the original and the reanalysis, there were no direction reversals: nothing that the original authors reported as going up appeared in the reanalysis as going down. Residual differences in the lists of significant features turned out to be attributable to threshold placement, tool-version differences, and preprocessing choices rather than to any discrepancy in the underlying quantities, which is precisely the kind of benign variation that experienced reanalysts expect.

Beyond reproducing single studies, the framework demonstrates capabilities that move toward genuine scientific synthesis. The agents were able to identify semantically similar studies—papers addressing comparable biological questions even when they use different vocabulary—and to judge whether datasets from different sources are compatible enough to be analyzed together. The capstone demonstration is a random-effects meta-analysis across multiple proteomics studies of liver fibrosis, in which the agents combined reanalyzed data from independent publications and recovered a consistent pattern of protein regulation. Meta-analysis is among the most valuable and most labor-intensive activities in biomedicine, and the prospect of automating the data-gathering and harmonization steps, while keeping the statistical model transparent and auditable, is among the most consequential implications of this work.

The authors are explicit that the study is a feasibility demonstration, not a validated benchmark of literature-wide performance. Five successful reanalyses, however carefully chosen, do not establish that the system would succeed on an unbiased sample of the literature, and the corpus-scale catalogue was not independently validated for accuracy. What the work does establish is an auditable, reusable toolset: the containers, the MCP-exposed tools, and the agent workflows are all designed to be inspected, rerun, and extended by others. That design philosophy matters because agentic systems introduce their own reproducibility risks. A language model’s behavior can vary across runs and model versions, and the authors’ decision to document every human intervention and workflow accommodation provides a template for how such systems should be reported if their results are to be trusted.

The broader significance lies in what it suggests about the future of data reuse. If the friction that keeps most published omics data unused can be reduced to a prompt, the effective size of the public data commons multiplies enormously. Individual laboratories could re-examine published datasets in light of new hypotheses without negotiating the plumbing of repositories and pipelines; meta-analyses could draw on far more studies; and negative results buried in supplementary files could be surfaced systematically. At the same time, the study’s honest accounting of human guidance serves as a reminder that the last mile of computational biology—judging whether a dataset truly fits a question, and whether a pipeline’s quirks matter for a given conclusion—still benefits from expert oversight. What Hutton and Meyer have shown is that the plumbing no longer has to be the hard part, and that the agents needed to do it are no longer hypothetical.

Subject of Research: Agent-supported retrieval, reanalysis, and synthesis of published omics data using large language model agents

Article Title: Omics data discovery agents: Agent-supported retrieval, reanalysis, and synthesis of published omics data

Article References: Hutton, A., & Meyer, J. G. (2026). Omics data discovery agents: Agent-supported retrieval, reanalysis, and synthesis of published omics data. PLOS Computational Biology, 22(10), e1014822. https://doi.org/10.1371/journal.pcbi.1014822

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014822

Keywords: omics, large language models, AI agents, reproducibility, proteomics, RNA-seq, meta-analysis, Model Context Protocol, data reuse, bioinformatics, PubMed Central, liver fibrosis

The Model Context Protocol, or MCP, is an emerging standard for connecting artificial intelligence assistants to external tools and data sources. Rather than allowing a language model to improvise its way through an analysis, an MCP server exposes a defined set of capabilities that the model can invoke in a structured way. stMCP applies this idea to spatial transcriptomics by pairing a natural language interface with a modular orchestration layer that routes requests to predefined analytical tools. When a researcher asks a question about their tissue data, the language model interprets the intent, the orchestration layer selects the appropriate tool, and the computation runs on the user’s own machine. The results are then returned to the user through a web-based viewer, accompanied by comprehensive analysis logs that document every step.

The distinction between orchestration and execution is central to the design. In many AI-driven analysis systems, autonomous agents generate and run code in extended back-and-forth exchanges with a language model, which can be slow, expensive, and difficult to reproduce. stMCP takes a more constrained approach. Because the analytical operations are performed locally by established tools rather than by code the model writes on the fly, biological data never needs to leave the user’s system, a significant consideration for laboratories handling sensitive or proprietary datasets. The architecture also reduces repeated language-model interaction during computation, meaning the model is consulted for interpretation and routing rather than for every computational step, which improves both efficiency and reproducibility.

Accessibility was a driving concern for the developers. Spatial transcriptomics platforms such as Xenium generate rich maps of gene activity within tissue, but extracting biological insight from those maps has largely been the province of computational biologists. stMCP includes a streamlined installation workflow that reduces the software configuration burden required for deployment, aiming to bring the technology within reach of experimental laboratories without dedicated programming support. Session management allows users to carry on iterative analyses across multiple interactions, building on previous results rather than restarting from scratch each time, and the analysis logging ensures that the full chain of decisions remains auditable long after the session ends.

Benchmarking results reported in the paper suggest that the constrained architecture is robust across the common tasks of spatial transcriptomics analysis. The system handled standard workflows reliably and, importantly, recovered gracefully from ambiguous or unsupported requests, situations that often derail less structured AI tools. The team also demonstrated that stMCP could successfully analyze larger public Xenium datasets and support workflows across multiple spatial transcriptomics platforms, indicating that the framework is not tied to a single technology or data format. This platform flexibility matters in a field where instruments from different manufacturers produce data with distinct characteristics and processing requirements.

Perhaps the most thought-provoking portion of the study is a direct comparison with STAgent and SpatialAgent, two autonomous agent-based systems for spatial transcriptomics analysis. The comparison revealed meaningful differences in execution efficiency, language-model utilization, reproducibility, and deployment requirements between the constrained MCP-based orchestration of stMCP and the more free-form agent-based approach. Autonomous agents promise flexibility, but that flexibility comes at a cost: more calls to the language model, greater computational overhead, and analyses that can be harder to reproduce because the model may take different paths on different runs. stMCP’s predefined tools and structured routing trade some of that flexibility for predictability, a trade-off the authors argue is well suited to scientific work, where reproducibility is not optional.

The implications extend beyond spatial transcriptomics. The authors frame stMCP as a demonstration of how MCP-based architecture can connect natural language interfaces with scientific computing tools generally, balancing accessibility, computational control, and reproducibility. Many scientific domains face the same challenge: powerful, well-validated analysis pipelines exist, but they are locked behind command-line interfaces and steep learning curves. A protocol-driven server that lets a language model orchestrate those pipelines, without letting it rewrite them, offers a template for democratizing computational science without sacrificing rigor. If the pattern holds across fields, the role of the AI assistant shifts from an unpredictable code generator to a reliable translator between human intent and established software.

The work emerged from a collaboration spanning Case Western Reserve University and Duke University School of Medicine, with contributions from departments of biomedical engineering, pharmacology, ophthalmology, and integrative immunobiology. The research was supported by funding from the National Institutes of Health, and the spatial transcriptomics data used in the study was generated with support from the CWRU Imaging Core. The authors note that generative artificial intelligence tools were used to assist with code debugging and language editing during manuscript preparation, while all analytical decisions, data interpretation, and final manuscript content were reviewed and verified by the authors, a transparency practice that mirrors the careful human oversight built into the software itself.

For the spatial transcriptomics community, stMCP arrives at a moment of rapid growth. As instruments scale up and public datasets expand, the bottleneck is shifting from data generation to data interpretation. Tools that lower the computational barrier without lowering scientific standards could determine how broadly the technology’s benefits spread, from large genomics centers to individual laboratories studying development, disease, and tissue architecture. By keeping data local, preserving established methods, and logging every analytical step, stMCP makes a case that the future of AI-assisted science lies not in replacing the computational toolkit but in making it conversational. The study is published open access, allowing researchers everywhere to examine, adopt, and build upon the framework.

Subject of Research: A Model Context Protocol server enabling natural language analysis of spatial transcriptomics data

Article Title: stMCP: spatial transcriptomics with a Model Context Protocol server

Article References: Smith, J. J., Wang, X., McPheeters, M., Widjaja-Adhi, M. A. K., Littleton, S., Saban, D. R., Golczak, M., & Jenkins, M. W. (2026). stMCP: spatial transcriptomics with a Model Context Protocol server. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06693-9

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06693-9

Keywords: spatial transcriptomics, Model Context Protocol, large language models, bioinformatics, Xenium, gene expression, reproducibility, AI agents, computational biology, open access software, natural language interface, BMC Bioinformatics

At the heart of the announcement is Scite, the citation-analysis platform acquired and operated by Research Solutions, and specifically the Scite MCP, a Model Context Protocol server that connects large language model assistants directly to indexed scholarly content. Under the agreement, MDPI’s journals are now fully indexed in Scite and delivered through the Scite MCP, which means that researchers working in Claude, ChatGPT, Microsoft Copilot, and other MCP-compatible tools can surface MDPI articles in their AI-driven workflows with proper attribution and direct links back to the Version of Record, the authoritative published version of each article hosted on MDPI’s own platform. For the first time, MDPI will also receive article-level usage reporting that reveals how its content is being accessed through AI channels.

The problem the partnership addresses is structural rather than cosmetic. When a researcher asks an AI assistant about a scientific topic today, the model may draw on published research that has been ingested during training or retrieved through a search layer, but the resulting answer frequently appears without consistent attribution, without a link to the authoritative article, and without any signal to the publisher that its content was used at all. Publishers, in other words, have had almost no visibility into how their journals perform inside AI tools, and authors have had little assurance that their findings are being represented accurately. This opacity matters in science, where the provenance of a claim is often as important as the claim itself.

The Scite Publisher Gateway is the infrastructure Research Solutions has built to solve these challenges. Once a publisher’s content is indexed through the gateway, that content can surface in AI responses enriched with citation context and direct links to verified article text. Attribution metrics are then reported back to the publisher through a dashboard, restoring a measure of the feedback loop that traditional web analytics provided in the era of browser-based discovery. The scale of adoption has been notable: since launching earlier this year, the Scite MCP has served more than 22 million requests, and monthly request volume has been growing at a rate exceeding 30 percent month over month, a figure that underscores how quickly agentic research workflows are moving from novelty to routine.

A central technical ingredient in the collaboration is Scite’s Smart Citations technology. Conventional citation counts treat every reference as an undifferentiated point of credit, but a citation in a review that endorses a finding is a very different signal from a citation in a paper that disputes it. Smart Citations classify and display the context in which one paper cites another, showing whether the citing work supports, contrasts with, or simply mentions the cited claim. When MDPI articles are surfaced through Scite, that contextual layer travels with them, allowing researchers using AI assistants to see not just that a paper exists but how its findings have been evaluated by the broader literature. This is a meaningful safeguard against the flattening of scientific nuance that can occur when complex evidence is compressed into a conversational answer.

Dr. Constanze Schelhorn, Chief Publishing Officer at MDPI, framed the move as a matter of grounding AI outputs in the scholarly record. As she put it, researchers are increasingly turning to AI assistants to navigate the literature, and it is important that AI-generated answers are grounded in that record. By making MDPI content available through Scite.ai and the Scite MCP, she noted, researchers can discover and engage with the publisher’s open access publications in AI-driven environments, augmented by Smart Citations and direct links to the version of record. Her comments reflect a broader conviction among open access advocates that accessibility alone is no longer sufficient; discoverability inside the tools where researchers actually work has become the new frontier.

The operational design of the partnership is also worth noting. Research Solutions handles the indexing, transformation, authentication, and integration required to bring MDPI’s portfolio into the Scite ecosystem, which means no technical lift is required from MDPI itself. That division of labor matters for publishers of every size, because building and maintaining MCP infrastructure, managing authentication for AI clients, and ensuring that retrieved text matches the published record are nontrivial engineering problems. By outsourcing that layer, MDPI can extend its distribution into AI channels without diverting resources from its core publishing operations, while retaining the direct connection between AI-surfaced content and the canonical article on its platform.

Sharon Mattern Büttiker, Vice President of Publisher Relations and Content Strategy at Research Solutions, emphasized that open access has always been about making research more accessible, but that as discovery moves into AI tools, publishers need confidence that researchers are finding the right source and that authors are receiving proper attribution. She described MDPI’s decision as an important step toward meeting researchers where they are while maintaining a direct connection to the authoritative version of its content and gaining greater visibility into how that content is discovered. Her remarks point to a recurring tension in the current moment: even openly licensed articles are, in many cases, blocked or restricted when AI agents attempt to retrieve them from the web, a paradox that undermines the very premise of open publishing.

Josh Nicholson, PhD, Chief Strategy Officer at Research Solutions, placed the agreement within the company’s wider agentic research toolkit and publisher strategy. The Scite Publisher Gateway, he explained, is one component of a strategy to help publishers bring research articles into AI. While the MDPI agreement primarily concerns open access articles, he observed that even these are now blocked by many publishers on the web, and he argued that the partnership will make content accessible and discoverable where the researchers of the future will live: in AI. The statement captures a bet that the industry is making collectively, namely that the next generation of literature discovery will be conversational and agentic, and that publishers who establish verified, attributable pipelines into those environments early will be best positioned to protect both their content and their revenue.

The practical consequences for researchers are immediate. A scientist using Claude, ChatGPT, or Microsoft Copilot can now retrieve MDPI articles through the Scite MCP with the confidence that the text returned is the verified published version rather than a preprint, a scraped copy, or a hallucinated reconstruction. Every surfaced article carries attribution and a link to the Version of Record, and citation context from Smart Citations helps the researcher judge how the community has received the work. For MDPI, article-level analytics will reveal which articles are accessed, by which organizations, and across which AI platforms, data that has been effectively invisible in the AI era until now. The full text of MDPI’s publication portfolio is already indexed and discoverable through the gateway, making this one of the most comprehensive integrations of an open access catalog into AI research tooling to date. As the volume of AI-mediated literature queries continues to climb, the partnership offers a template for how publishers, authors, and AI platforms can align their incentives: researchers get grounded answers, authors get credit, and publishers get the measurement they need to sustain the open access model.

Subject of Research: Integration of an open access publisher's journal portfolio into AI research tools via a citation-aware indexing gateway

Article Title: MDPI and Research Solutions’ Scite collaborate to make open access research discoverable, attributed, and measurable inside AI tools

Article References: MDPI and Research Solutions’ Scite collaborate to make open access research discoverable, attributed, and measurable inside AI tools. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: MDPI, Research Solutions, Scite, open access, Model Context Protocol, AI assistants, Smart Citations, scholarly publishing, citation context, Version of Record, publishing analytics, research discovery

The open-source EnergyPlus engine, developed by the Department of Energy, performs the underlying calculations that predict how a design will perform. Most tools operate on its text input files, but practitioners typically work one layer up, in the OpenStudio software development kit, which provides a typed model structure and a library of reusable transformation scripts known as Measures. The new server deliberately builds on that SDK layer rather than raw input files, because the typed object structure is more composable and less error-prone to manipulate, and it connects AI agents to the broader ecosystems of OpenStudio-standards, ComStock, and the Building Component Library. Earlier protocol-based servers, such as EnergyPlus-MCP, work only at the input-file layer and defer geometry creation and HVAC loop construction to future work; OpenStudio-MCP tackles the full lifecycle from creation through simulation and evaluation.

The technical heart of the system is the Model Context Protocol, introduced by Anthropic in late 2024, which allows a language model to drive external software by calling validated, schema-typed tools instead of emitting free-form code. OpenStudio-MCP exposes 197 such tools, organized into self-contained skill modules that are discovered automatically at startup. They span model creation, geometry and thermal zoning, construction and schedule assignment, HVAC synthesis, simulation control, results extraction, quality assurance, and even large-scale parametric analysis on OpenStudio-server. A single call to add_baseline_system, for example, wires a complete air- and plant-loop topology corresponding to one of the ten baseline system types defined in ASHRAE Standard 90.1 Appendix G, a task that has historically required careful manual scripting.

Because language models can call operations out of order, select unsuitable systems, or simply invent SDK methods that do not exist, the server wraps the SDK at two levels. Low-level tools expose explicit OpenStudio operations as typed calls, while higher-level tools encode common workflows such as whole-building creation, baseline HVAC assignment, and result extraction. A separate knowledge layer serves curated workflow guides covering object dependencies, ASHRAE system-selection rules, and Measure authoring, which agents retrieve on demand. The server also guards against the finite context window of any language model: instead of dumping a 5,000-space model or a full 8,760-hour annual result into the conversation, it returns compact structured summaries, previews models without loading them, and distills entire simulations into a handful of summary numbers while the model files themselves remain server-side.

Perhaps the most striking capability is automated Measure authoring. Measures are programs, and extending analysis beyond the off-the-shelf library has traditionally required engineers to double as software developers, putting custom analysis out of reach for many architects and engineers who best understand the building. With OpenStudio-MCP, an agent scaffolds a new Measure from a plain-language request, writes its logic, runs its tests, and applies it inside a sandboxed environment that runs child processes as unprivileged users under a fail-closed Landlock filesystem policy, a seccomp filter denying outbound network access, and strict resource limits. The authors stress these are implemented controls rather than security guarantees, since the software has not undergone a formal external penetration test, but targeted adversarial probes using canary listeners and decoy secrets exercised cross-tenant file reads, environment-secret capture, filesystem escapes, and forged transfer requests.

To demonstrate the system end to end, the researchers gave an AI agent a natural-language request specifying a medium office building, its location, a baseline HVAC system, a comfort criterion, and a four-pipe active-chilled-beam retrofit, naming no tools. Working from an empty session with Claude Opus 4.8, the agent created and simulated a 27-zone, three-story, 53,600-square-foot office using the ASHRAE 90.1-2019 template with a variable-air-volume reheat system. The initial model narrowly failed the comfort criterion with 301.7 occupied unmet hours. Digging into sizing reports, the agent found that heating capacity was adequate but a 50 percent reheat-mode airflow cap was constraining morning warm-up after nighttime setback, raised the cap on all 27 terminals, and brought the model to 78.0 unmet hours at a site energy use intensity of 42.4 kBtu per square foot, squarely within the middle half of the observed U.S. office stock from the 2018 Commercial Buildings Energy Consumption Survey.

The agent then authored a Measure that replaced each terminal with a four-pipe beam connected to the existing chilled- and hot-water plants, verified SDK class names, passed its own tests, and validated 27 beam terminals with no errors, all without a human writing or reviewing code. The retrofit maintained comfort but increased site energy by 8.7 percent, driven by a 179 percent surge in fan energy, because the constant-volume beams forfeited the variable-volume air handler’s part-load fan savings and economizer hours. Crucially, the agent reported this adverse result, explained both mechanisms, and proposed remedies including a right-sized dedicated outdoor air system with energy recovery. The session used 98 calls to 36 tools, three annual simulations, and roughly 21 minutes, with no human intervention after the initial request.

Feasibility is not reliability, so the team built a reproducible benchmark of 16 graded tasks across six families, tested with Claude Opus 4.8, Opus 4.6, Sonnet 4.6, and Haiku 4.5 alongside GPT-5.4 and GPT-5.4-mini. Each trial was graded by two deterministic checks without any AI judging: whether the agent called an acceptable tool, and whether the saved model passed physical checks such as assembly R-values, HVAC loop membership, and pinned EnergyPlus outputs. The distinction proved essential, because in 18 of 23 outcome failures an agent replaced a roof assembly with one up to 1.86 square meters kelvin per watt worse while reporting success, an error visible only in the saved artifact, not in the agent’s confident report. Under a common configuration with all tool schemas loaded, GPT-5.4 and Opus 4.8 achieved 100 percent outcome rates, with Opus 4.6 at 95.8 percent, GPT-5.4-mini at 93.8, Sonnet at 91.7, and Haiku at 85.4.

The ablation results carry a practical lesson for anyone deploying agentic AI. Loading every tool schema up front raised the weakest model’s success rate by 12.5 points but inflated costs by 36 to 68 percent for stronger Claude tiers while changing outcomes by at most two tasks, meaning deferred schema discovery saves money for capable models at no accuracy loss. The curated knowledge layer, surprisingly, changed no model’s completion rate by more than 6.3 points. The completion budget also mattered: at a 120-second limit, 17 of Opus 4.8’s 18 failures were timeouts, yet it passed every trial in four of five configurations when given 600 seconds, showing that slow but productive work was being misclassified as failure. An unscaffolded baseline without the server showed agents can handle basic OpenStudio operations through direct scripting, but both tested models failed a task requiring exact counting of warnings in an EnergyPlus error file.

The researchers frame the work as broadening access rather than replacing rigor. Engineers and architects could request, test, and run bespoke retrofit analyses without writing Ruby, while organizations could offer shared modeling capacity to design firms, classrooms, or utility programs by issuing authentication tokens instead of provisioning workstations, turning energy modeling from a per-seat desktop activity into shared infrastructure. Validation and quality-assurance tools, audit records of every tool call, and artifact-based grading make the checking explicit, but the authors are careful to note that engineering judgment is not automated: generated artifacts and conclusions require review by a qualified practitioner before use in real engineering decisions. The code, benchmark harness, and archived trial records are openly available under a BSD-3-Clause-style license, inviting the building science community to put AI-driven modeling to the test.

Subject of Research: An open-source Model Context Protocol server enabling AI agents to perform full-lifecycle building energy modeling with the OpenStudio SDK.

Article Title: OpenStudio-MCP: a model context protocol (MCP) server for AI agent-driven building energy modeling with the OpenStudio SDK

Article References: Ball, B. L., Long, N., Fleming, K., & Goldwasser, D. (2026). OpenStudio-MCP: a model context protocol (MCP) server for AI agent-driven building energy modeling with the OpenStudio SDK. SoftwareX, 36, Article 103020. https://doi.org/10.1016/j.softx.2026.103020

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103020

Keywords: OpenStudio-MCP, building energy modeling, Model Context Protocol, large language models, AI agents, EnergyPlus, OpenStudio SDK, HVAC synthesis, Measure authoring, ASHRAE 90.1, agent benchmark, sandboxing

The journey begins in New York, where Founder and Chief Executive Officer Dr. Alex Zhavoronkov will attend the Morgan Stanley 24th Annual Global Healthcare Conference from September 14 to 16. On September 15 at 14:35, Zhavoronkov is scheduled to participate in an in-person fireside chat, engaging global investors and industry leaders on the company’s latest advances in generative AI-driven drug discovery, aging clocks and anti-aging interventions. The Morgan Stanley conference is widely regarded as one of the largest and most influential healthcare investment gatherings in the world, convening thousands of leaders each year, from multinational pharmaceutical companies and biotech innovators to medical device makers, digital health pioneers, hedge funds, long-only investors, consulting firms and regulatory bodies. Its mix of keynote addresses, fireside chats, one-on-one investor meetings and forward-looking roundtables makes it a core venue where international capital identifies healthcare opportunities and where large pharmaceutical companies scout innovative technologies and potential acquisition targets.

For Zhavoronkov, the New York appearance is an opportunity to present Insilico’s progress to the capital markets at a moment of unusual momentum. The company has been steadily expanding its narrative beyond a single headline asset, highlighting an end-to-end autonomous laboratory roadmap that pairs its generative chemistry platforms with laboratory automation designed to compress the timelines of target identification, molecular design and preclinical validation. The aging research dimension of the company’s work, including its well-known deep learning aging clocks that estimate biological age from multimodal data, has long differentiated Insilico from AI drug discovery peers, and executives are expected to weave that longevity science perspective into their dialogue with investors who increasingly view aging biology as a fertile ground for new therapeutics.

From New York the focus shifts to the Middle East. Dr. Alex Aliper, Co-Founder and President of Insilico Medicine, has been invited to the Riyadh Global Medical Biotechnology Summit, known as RGMBS 2026, running September 14 to 16 in the Saudi capital. On September 16, from 09:00 to 12:00 at the Sofitel Riyadh Hotel and Convention Center, Aliper will lead the Insilico team in hosting a hands-on workshop titled Model Context Protocol-Empowered Protein Design: Combining AI Foundation Models and Physics-Based Molecular Modelling. The session is designed to be intensely practical. Participants will gain first-hand experience using the Model Context Protocol, or MCP, to connect AI foundation models, molecular simulation engines and chemical databases. The curriculum covers the complete workflow from protein and ligand structure preparation through physics-based validation, teaching attendees how to score and prioritize drug candidates with AlphaFold, RDKit and OpenMM, how to interpret binding modes, kinetics and free-energy calculation results, and how to run a directed MCP workflow inside Insilico’s Chemistry42 sandbox environment.

The choice of technical material is significant. The Model Context Protocol has emerged as an open standard for connecting large AI models with external tools and data sources, and Insilico’s workshop represents one of the most concrete demonstrations of how that architecture can be applied to protein engineering and small-molecule discovery. By linking generative foundation models to physics-based simulation, the workflow aims to marry the speed and creativity of deep learning with the rigor of molecular mechanics, free-energy perturbation and kinetics analysis that medicinal chemists have long demanded. Aliper, who has spent much of his career at the intersection of AI-driven drug discovery, frontier biomedical science and cross-disciplinary tool integration, will also use the summit to showcase Insilico’s role in building the Chemistry42 generative chemistry platform and the company’s broader autonomous laboratory ecosystem, positioning the workshop as a window into how modern AI-native biopharma companies orchestrate computational and experimental workflows.

RGMBS 2026 itself carries strategic weight. Co-initiated by the Saudi Ministry of Health, the Royal Commission for Riyadh City and leading biomedical authorities, the summit is one of the largest international biotechnology and medical innovation gatherings in the Middle East. It convenes scientists, research and development leaders, clinical experts, regulators and strategic investors spanning biopharmaceuticals, gene and cell therapy, medical devices, digital health and fundamental life sciences. Organizers have centered the program on frontier biotechnology, precision medicine, AI-driven drug discovery, translational medicine, health-tech investment and biomanufacturing, using keynote addresses, themed workshops, closed-door sessions, industry matchmaking and project roadshows to drive cross-regional collaboration. For global biopharma companies, the event is increasingly viewed as a gateway to the Middle East market and to Saudi Vision 2030, the kingdom’s national strategy that places biomedical capability among its economic diversification priorities.

The third stop brings Insilico to Shanghai, where Co-CEO and Chief Scientific Officer Dr. Feng Ren will attend Bio-Shanghai Week 2026, an event anchored by Zhangjiang Drug Valley, a national-level biopharmaceutical industry hub. On September 17 at 15:00, during the opening ceremony’s AI-Driven Innovation session, Ren will engage in an in-depth dialogue with Professor Michael Levitt, the 2013 Nobel Laureate in Chemistry and Stanford University structural biology professor, on the theme of AI-driven global innovation in therapeutic target discovery and treatment technologies. The conversation is expected to traverse three dimensions: foundational science breakthroughs, industrial translation pathways and global strategic coordination, examining how artificial intelligence is systematically reshaping the full chain from target discovery through molecular design to clinical development. Ren, regarded as one of the leading scientists driving AI-enabled drug research and clinical translation in China, will share Insilico’s generative AI platform, its pipeline progress and the company’s global footprint, creating what organizers describe as a high-level exchange between a leading Chinese AI-driven pharmaceutical company and a top global scientist.

Bio-Shanghai Week ranks among the largest and most internationally influential biopharmaceutical industry events in Shanghai, drawing leading scientists, clinical experts, multinational pharmaceutical and biotech companies, innovative drug and device developers, investors, regulators and industry service platforms. Its agenda spans AI-driven innovation, gene and cell therapy, antibodies and antibody-drug conjugates, rare diseases, neuroscience, global market access, clinical translation and the broader industry ecosystem. The event serves as a vital window into the frontier of China’s biopharmaceutical industry and the wider Yangtze River Delta innovation ecosystem, a region that has become one of the world’s densest concentrations of drug discovery talent and capital.

The final leg of the tour takes Zhavoronkov to Boston on September 18 for the Harvard IvyTech Discussion, co-initiated by Harvard University and other Ivy League academic institutions. From 10:50 to 12:00, he will deliver a keynote address in a forum titled A Geo-Economic Shift: China’s Rise as an Innovation Powerhouse in Biotech. Sharing the stage with leading scientists, industry strategists and policy researchers from North America’s top institutions, Zhavoronkov will discuss the leapfrog transformation of China’s biopharmaceutical industry from generic manufacturing to first-in-class innovation, and the corresponding evolution of the global biopharma value chain and capital landscape. He is also expected to present Insilico’s strategic blueprint as what the company calls a bridge enterprise connecting Eastern and Western innovation ecosystems, spanning Chinese foundational research, the company’s AI platform technology, its global research and development pipeline, and its international capital and industry partnerships. The IvyTech platform, which focuses on frontier technology, industrial transformation and geo-economic topics, brings together scientists, technology executives, entrepreneurs, policymakers and institutional investors for dialogue across biomedical innovation, artificial intelligence, advanced manufacturing, the energy transition and cross-border innovation ecosystems.

The speaking tour arrives at a pivotal financial and scientific moment for Insilico Medicine. The company recently reported total revenue of approximately 106 million US dollars in the first half of 2026, a 287 percent year-over-year increase, and achieved its first profitable half-year since listing, with adjusted net profit exceeding 51 million dollars. The milestone was driven by a series of out-licensing, co-development and research collaborations with global partners including Eli Lilly, Servier, Takeda, SK Biopharmaceuticals, Qilu Pharmaceutical, Hygtia Therapeutics, CMS and Tenacia. As of the latest practicable date, the total contract value of transactions announced by the company in 2026 reached approximately 7.3 billion dollars, pushing the cumulative contract value of its major collaborations since 2021 to roughly 11 billion dollars. On the research front, Insilico nominated nine development candidates within the first nine months of 2026 as of late August, a company record for annual pipeline productivity, and achieved eight clinical milestones across its proprietary and co-developed programs. Leading that progress is rentosertib, also known as ISM001-055, the world’s first drug candidate discovered and developed using generative AI, which has advanced into a Phase III trial evaluating treatment for idiopathic pulmonary fibrosis, a progressive and often fatal scarring lung disease with few therapeutic options. Listed on the Main Board of the Hong Kong Stock Exchange on December 30, 2025 under the stock code 03696.HK, Insilico continues to apply its Pharma.AI platform to fibrosis, oncology, immunology, pain, obesity and metabolic disorders, while extending the technology into advanced materials, agriculture, nutritional products and veterinary medicine. The four-city executive tour, spanning capital markets in New York, biotechnology diplomacy in Riyadh, scientific dialogue in Shanghai and academic strategy in Boston, functions as a compressed portrait of the company’s thesis: that generative AI, rigorous physics-based validation and global collaboration can deliver better drugs faster, and that the companies able to bridge the world’s major innovation hubs will define the next decade of biopharmaceutical progress.

Subject of Research: Insilico Medicine executive participation in four global healthcare and AI summits showcasing generative AI drug discovery

Article Title: Across four global innovation hubs: Insilico Medicine executive team to speak at premier healthcare and AI Summits

Article References: Across four global innovation hubs: Insilico Medicine executive team to speak at premier healthcare and AI Summits. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Insilico Medicine, generative AI, drug discovery, rentosertib, AlphaFold, Model Context Protocol, aging clocks, biotechnology, idiopathic pulmonary fibrosis, AI summits, autonomous laboratory, biopharmaceutical innovation