BioRouter AI Agent Marketplace
A curated, quality-controlled ecosystem of AI agents for biomedical research. Install any agent directly into BioRouter via the Model Context Protocol.
An MCP server enabling AI-driven SQL queries on the UCSF OMOP de-identified clinical database — read-only access to standardized electronic health records organized under the OMOP Common Data Model. Enables natural language querying of real-world clinical data.
An MCP server for querying the SPOKE biomedical knowledge graph — a large-scale graph database linking diseases, genes, proteins, drugs, and biological pathways. Execute Cypher queries for rapid biomedical knowledge inference and literature-grounded hypothesis generation.
Medical Model Context Protocol — transforms BioRouter into a medical AI assistant by providing secure, local access to electronic health records and biomedical knowledge graphs. Process sensitive health data and deliver clinical insights without transmitting data to the cloud.
An MCP server that gives BioRouter full browser automation capabilities via Playwright. Navigate websites, extract data, fill forms, and automate web-based research workflows using structured accessibility snapshots — no vision model required.
More agents are being added. Contribute on GitHub
Federated Workflows
Community Recipes
Shareable, AI-adapted research pipelines — package a workflow once, run it across institutions.
Query an OMOP-based EHR to identify a Type 2 Diabetes patient cohort, compute demographic summaries (gender, age distribution, race, ethnicity), and automatically generate a shareable HTML report and an interactive R Shiny dashboard — all via natural language, runnable at any OMOP-compatible institution without sharing patient data.
version: 1.0.0 title: EHR Diabetes Demographics and Reporting Dashboard description: A recipe for querying an OMOP-based EHR to derive a disease cohort (e.g., T2D), summarize demographics, and produce sharable visual outputs. instructions: Use an OMOP CDM clinical database to identify a patient cohort via diagnosis concepts (e.g., matching condition_occurrence to relevant concepts in concept). Then, join to person to obtain demographics (gender, age via year_of_birth, race, ethnicity) and compute counts and percentages. Present results as both a concise HTML report and, optionally, an interactive R Shiny dashboard with charts and tables. When tooling is available, run SELECT-only queries through a clinical data extension (e.g., medcp.query_clinical_records), and use a system automation/file tool (e.g., computercontroller.automation_script or computercontroller.computer_control) to generate files on disk (HTML and an R Shiny app folder). Make output locations explicit (e.g., /Users/<user>/Desktop) and ensure visualizations include clear labels and percentages. extensions: [] activities: - Count disease cohort - Demographic breakdowns - Generate HTML report - Build Shiny dashboard - Age distribution analysis parameters: []
How to Install an Agent
Copy the Command
Click the copy icon on an agent card to copy its install command to your clipboard.
Add to BioRouter
In BioRouter: Sidebar → Extensions → Add custom extension → paste the command.
Enable & Use
Toggle the extension on. The agent's tools are now available in your sessions.
How to Execute a Recipe
Download the Recipe
Click Download Recipe on any recipe card above to save the .yaml file to your machine.
Open BioRouter Recipes Tab
In BioRouter, navigate to the Recipes tab in the sidebar.
Import Recipe
Click Import Recipe and select the downloaded .yaml file. BioRouter will configure the workflow automatically.
Run & Share
Execute against your institutional data. Share the YAML with collaborators — they run it on their own data, no patient records exchanged.
Want to create or share a recipe with the community? Contact Wanjun Gu for more info.
Have a biomedical AI agent to contribute? Submit it to the BAAM marketplace.
Contribute on GitHub