MCP Server for Trieve to support usage of Trieve in agentic workflows from Model Context Protocol clients. Ideally, this is built in a way where it's really exposing the set of tools/prompts/resources that could be consumed, with MCP as one possible shape.
Offering tools enables agents to intelligently consume Trieve's APIs.
I think an initial version of Trieve would look like:
- MCP server with create-mcp-server (there are multiple starters) connecting to Claude over stdio json-rpc. SSE exists but can be added later and is only recommended if the mechanics are actually needed between the client and the server
- Add config for Trieve's SDK lib
- Tools available that map Trieve's OpenAPI specification https://api.trieve.ai/redoc for descriptions oriented for LLMs, then wire up the tool calls to the SDK. These are things that the user has authority over executing for
- Resources that map to Trieve's API, but for anything that the user would potentially want to bring into context (in the attach a file sense), NOT call out and get a tool. These don't pass through the LLM, and can be used for bringing in files for datasets, etc.
Initially, just having the pipeline of - I'm in Claude Desktop, and I want to call a tool to hit the Chunk endpoint and get the results, that's a good springboard project to continue fleshing out an agent.
- New MCP server (TS and Python have official clients, but it's a separate process)
Matrix is preferred. Reach out on discord or Matrix for further assistance.
