trapi-mcp

# TRAPI MCP A FastMCP-based tool for writing prompts using Translator Reasoner API (TRAPI) queries. This package enables easy integration of biomedical knowledge graph queries into your LLM workflows. ## Installation You can install the package from source: ```bash pip install -e . ``` Or using uv (recommended): ```bash uv install trapi-mcp ``` ### Python API ```python from trapi_mcp.tools import trapi, trapi_status, trapi_results # Submit a TRAPI query response = trapi( subject="MONDO:0005148", # Alzheimer's disease object_="CHEBI:6801", # Acetylcholine predicate="biolink:affects" ) # Get the query ID pk = response["pk"] # Check status status_response = trapi_status(pk) print(f"Query status: {status_response.get('status')}") # When status is "Done", get results if status_response.get("status") == "Done": results = trapi_results(pk) # Process results ``` ## Developer Guide ### Setup Development Environment 1. Clone the repository: ```bash git clone https://github.com/your-username/trapi-mcp.git cd trapi-mcp ``` 2. Install development dependencies: ```bash uv install ``` ### Architecture The package consists of several key components: - `api_utilities.py`: Low-level functions for interacting with Translator API services - e.g., how to make API calls - `tools.py`: High-level functions for building and executing TRAPI queries. Currently, NameResolver, NodeNormalizer, and ARS query endpoints are all callable as MCP servers from this repository. - `main.py`: FastMCP integration and CLI setup ## Integration with Goose (Example) [Goose](https://github.com/GSK-AI/goose) is a framework for building LLM applications. Here's how to integrate TRAPI MCP with Goose: #### install goose: https://block.github.io/goose/docs/getting-started/installation/ #### Set an LLM Provider https://block.github.io/goose/docs/getting-started/installation/#set-llm-provider #### Add TRAPI MCP as an extenstion https://block.github.io/goose/doc…