PhenoWorks MCP
PhenoWorks MCP (phenoworks-mcp) connects compatible AI assistants to the
PhenoWorks API through the Model Context Protocol. It gives them tools to find
data, inspect methods, run pipelines, and retrieve results using the caller's
access permissions.
The PhenoWorks Agent provides the conversational experience. MCP provides the tools that connect an assistant to the research workspace.
Available tools
| Task | Tools |
|---|---|
| Check the connection and identity | phenoworks_health, phenoworks_me |
| Find research data | list_projects, get_project, list_datasets, get_dataset, get_dataset_summary |
| Inspect analysis methods | search_analysis_blocks, describe_analysis_block, review_analysis_block_code |
| Check pipeline compatibility | validate_pipeline_steps |
| Run and monitor analysis | run_dataset_pipeline, list_pipeline_jobs, get_pipeline_job, cancel_pipeline_job |
| Retrieve outputs | list_artifacts, download_artifact |
Block descriptions and source review preserve the backend's permissions; those detail routes currently require an administrator. Source review returns stored code for inspection without executing it.
Connections and authentication
Use Streamable HTTP when the MCP server runs as a service, or stdio when a local client launches it as a process. Choose one credential mode: an API key, a bearer auth token, or per-request credential forwarding over HTTP.
For the integrated agent, a shared HTTP server can forward each user's
Authorization or X-API-Key header. An external local client can instead use
a dedicated API key. Access checks still happen in the PhenoWorks backend.
From a request to a result
The assistant finds a dataset, checks the available blocks and input requirements,
and submits a compatible pipeline. The returned pipeline_id identifies the
analysis run; operation_id identifies the queued operation. Use the pipeline
ID when checking jobs or retrieving their artifacts.
Clients that support form elicitation can present a confirmation before a run. Clients without that capability can submit directly, so a headless integration should handle its own approval policy.
Retrieval scope
MCP retrieves information through the tools listed above. Its dataset summaries and artifact access support the Scientific Knowledge Base, but do not automatically index every image, PDF, or manuscript in the workspace.
Follow Connect an MCP Client for setup and a first connection check.