The open-source LabMCP catalog exposes 411 actions, but K-Dense says none of its simulated connectors has been confirmed on a physical instrument.
By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)
· Published
Primary source: [K-Dense on X](https://x.com/k_dense_ai/status/2103984320457347339)
Why it matters #
LabMCP makes instrument-control software available for free, but its 32 connectors have not been confirmed on physical equipment. For labs, the decisive measure will be whether owners can verify the connectors on their instruments and configurations.
K-Dense, co-founded by Timothy Kassis, has released LabMCP, an open-source connector layer for AI assistants to operate laboratory instruments, with a limitation that puts its headline catalog in perspective: the software has been tested in simulators, not confirmed on physical equipment. For labs considering putting AI agents in control of equipment, that is the gap between an available interface and a demonstrated deployment.
K-Dense says all 32 LabMCP connectors remain simulated: they have been tested against simulators, but users have not yet confirmed them on physical instruments. The connectors expose 411 actions across instruments and protocols used in biology, chemistry, physics, health and engineering. The project is free and licensed under Apache 2.0; it was released on September 26th.
A simulator can exercise software against expected instrument responses. It cannot show that a connector works with a particular physical instrument and its configuration. K-Dense's repository says connectors are built from manufacturers' manuals and remain simulated until an instrument owner confirms them. K-Dense also says LabMCP is not validated for GxP or clinical use. The action count describes what the software offers, not 411 verified operations in working labs.
That validation work is the project's immediate deployment test. LabMCP runs on a computer connected to the instrument and gives an AI application a defined menu of actions through the Model Context Protocol (MCP). Supported clients include Claude Desktop, Claude Code, Cursor, VS Code, Codex and Windsurf. The repository offers simulator and read-only modes, configurable limits, prompts for hazardous actions and command logging. K-Dense says labs should retain instrument interlocks and other built-in protections; the software is not a certified safety system.
The local connector does not guarantee that experimental data stays local. LabMCP talks to instruments on the attached computer, while the AI application may send results to its model provider. That leaves labs responsible for checking their own data policies before connecting sensitive experiments.
A connector layer, not a complete lab system
LabMCP's architecture is aimed at existing equipment: it provides separate connectors for instrument families and general interfaces such as SCPI, Modbus, EPICS and SiLA 2. That differs from products that combine instrument access with broader lab software. Carbon describes a single edge daemon for instrument control across protocols including SCPI and Modbus. UniteLabs describes local instrument connectors linked through its GroundControl edge gateway, with SiLA 2 interfaces. Those are different routes to connecting equipment; the public materials reviewed here do not establish comparable physical-hardware validation results.
Scispot's MCP product is a different kind of alternative: its stated focus is conversational access to lab records and workflows such as experiments, samples and protocols, rather than direct control of physical instruments. The comparison is useful because an MCP connection can mean anything from querying a lab notebook to sending a command to a device. LabMCP is explicitly attempting the latter, with its current verification caveat attached.
K-Dense co-founder and CTO Timothy Kassis has worked across AI and bioengineering. Biostate AI lists him as its former Head of AI and Data from 2021 to 2025; Kassis has said he helped build Biostate's AI Scientist before that work spun out into K-Dense. He completed graduate work in bioengineering at Georgia Tech and later held a postdoctoral appointment at MIT, where he also served as a lead instructor for the Living Machines program (Georgia Tech; MIT). That background connects the connector project to K-Dense's broader effort to build AI tools for scientific work.
K-Dense lists its headquarters in Palo Alto. Accel's March 2026 AI Cohort listing identifies Kassis and David K. Zhang as founders and says K-Dense was selected for the Accel Atoms cohort, backed by Google's AI Futures Fund. The amount, instrument, valuation and terms of K-Dense's investment have not been disclosed. Biostate AI's financing is separate and should not be read as K-Dense funding.
LabMCP itself is free; users may need a separate subscription for their AI application. K-Dense's paid K-Dense Web product lists Plus at $199 per month, Team at $499 per month and Enterprise at custom pricing (K-Dense pricing). K-Dense says its separate open-source Scientific Agent Skills have been adopted by thousands, but it has not published a LabMCP user or deployment count. The repository showed eight stars and no forks in the supplied snapshot, an early signal rather than evidence of lab adoption.
K-Dense is asking instrument owners to run the connectors and report whether they work. Until those confirmations arrive, LabMCP is a free, inspectable starting point for connecting AI applications to lab equipment, not proof that its broad catalog is ready for routine physical use.