Biological AI requires an agent infrastructure: systems where agents can store, process, and visualize large molecular data sets from environments scientists already use. Biological tasks downstream of large local, single-cell, or proteomics experiments require agents to send calls to bioinformatics tools and parse results into text and plots in a conversational interface.
These tools often require large computing resources, so deployment must be reliable, especially as the volumes of data from modern testing increase. Run time and costs can quickly increase from trying and wrong results, especially from tool calls that take hours to days. They should also be processed with appropriate parameters and agent-readable documentation, duly provided by the original test developer, to facilitate correct use in all complex scientific situations.
At Latch, we’ve seen customers of our Solution Provider partners, such as TakaraBio, Vizgen, and AtlasXOmics, use both Latch Agent and external tools like Claude Code or Cursor to accelerate their data analysis. While Latch users can already use our SDK to programmatically interact with the platform, this limits interaction to desktop environments with a default shell environment.
Today we present the program Sleep MCPa remote MCP server that allows you to securely connect your favorite agent cables to the Latch platform and use certified bioinformatics tools built and maintained by kit and equipment suppliers. Agents can navigate through data on the Latch platform and run existing deployments of proven bioinformatics workflows for data analysis.
Latch MCP is also available with Directory of Claude connectors – Anthropic’s certified and updated catalog of MCP servers – allowing you to use Latch MCP on all Claude products, including Claude ScienceAnthropic’s newly launched AI workbench for scientists.
To show how scientists can use Latch MCP, we will go through an example: we analyze a Takara Trekker spatial pattern learning experiment in the rat brain. Trekker gives power the truth single-cell location mapping by directly distributing location barcodes to individual cell nuclei, eliminating the need for post-processing to resolve cell identity.
A single test can generate several gigabytes of data per sample. As the number of samples increases, it quickly exceeds the limits of a scientist’s personal computer. Furthermore, configuring a bioinformatics pipeline to take crude FASTQ samples all the way to purified h5AD is not a trivial task. With the Latch MCP server at Claude, scientists can simply point to their dataset and ask in plain language: “Run the Trekker pipeline workflow on my U_C mouse dataset.” Claude can access the pre-configured, gold-standard pipelines published in Latch by Takara’s organic experts and start the workflow directly. Scientists can follow the Latch Workflow link to monitor progress or ask Claude to track the status of the workflow itself.
Once the pipeline is complete, scientists can continue analysis using the Latch Agent. For example, the video above shows using the h5AD viewer embedded in the Latch Agent to visualize a brain dataset of a large Trekker mouse. Alternatively, scientists can ask Claude to do more analysis. Claude can access Latch Data, a cloud-based file system, to find relevant results files. From there, Claude or any other agent can use a rich library of proven local workflows:
-
NVIDIA RAPIDS for GPU-accelerated pre-processing
-
Cell2location or RCTD for cell-type deconvolution
-
BANKSY to combine space
-
And more
You can now use Latch MCP with any AI agent that supports MCP integration (setup instructions here).
