knowledgebase

BioContextAI Knowledgebase MCP provides a unified access layer to a suite of biomedical resources via the MCP protocol. It exposes tools for antibody registry lookups, literature search and metadata retrieval, gene and protein information, pathway and interaction data, and single-cell markers, among others. Agents and IDEs can connect to the MCP server using uvx (or your preferred MCP client configuration) to query these resources in a standardized way, enabling LLMs to retrieve verified biomedical information during reasoning and task execution. The server integrates multiple external APIs (e.g., Ensembl, EuropePMC, OpenTargets, Reactome, STRING, AlphaFold DB, etc.) and can be extended with OpenAPI-backed endpoints via a configurable config.yaml. When using the Knowledgebase MCP, adhere to each data source’s rate limits and licensing terms, and cite both BioContextAI and the source data when used in research or publications.

To use the MCP server, start the service with your preferred setup (local uvx, IDE integration, or Docker). Once running, you can issue queries that resolve to resource lookups such as gene IDs, literature articles, protein information, pathways, and cell-type markers. The server acts as a browser-like layer for your LLM, returning structured responses that your agents can parse and reason over, while keeping data sources decoupled from client logic.

Prerequisites:\n- A supported runtime (Python with uv) or Docker if you prefer containerized deployment. For local development with uvx, you should have uv installed and accessible on your PATH.\n- Basic tooling: Python (3.11+ recommended for compatibility), pip, and access to the internet to install the package.\n- Optional: Docker if you want to run a containerized deployment.\n\nInstallation steps (preferred local setup using uvx):\n1) Install the Knowledgebase MCP package from PyPI:\nbash\npip install biocontext_kb\n\n2) Start the MCP server via uvx:\nbash\nexport MCP_ENVIRONMENT=PRODUCTION && export PORT=8000 && uvx biocontext_kb@latest\n\n3) Verify the server is running by hitting the MCP endpoint (e.g., http://localhost:8000/mcp/).\n\nAlternative: Local development with IDE integration (e.g., VS Code or other editors):\n- Ensure your editor’s MCP config points to the uvx command and the package name, for example:\njsonc\n"mcpServers": {\n "biocontext_kb": {\n "command": "uvx",\n "args": ["biocontext_kb@latest"]\n }\n}\n\n\nDocker (containerized deployment):\n1) Build and run the container (example):\nbash\ndocker build -t biocontext_kb:latest .\ndocker run -p 127.0.0.1:8000:8000 biocontext_kb:latest\n\nThis will expose the MCP endpoint at http://127.0.0.1:8000/mcp/. Ensure proper security measures for public deployments.

Tips and notes:\n- OpenAPI extension: If you add OpenAPI-based services, place their schemas in src/biocontext_kb/openapi/config.yaml. By default, there are no extra OpenAPI servers; you can enable and configure them as needed.\n- Local hosting is recommended to avoid remote rate limits and to ensure availability for dependent agents.\n- If you are using Claude Desktop or other clients, you can configure local MCP servers in claude_desktop_config.json (see examples in the README).\n- When running publicly, consider reverse proxying (Nginx/Caddy), restricting ports, and running in a least-privilege container/user.\n- The data accessed through the MCP server is governed by source licenses; cite BioContextAI and the respective data source when used in research.