Support MCP to add specific document context (rather than just web resources)

[rlancemartin]

[bartolli]

Two ideas come to my mind

Connect the research agent to clinical notes database (synthetic demo data):

It connects live research findings to clinical notes from a local database, like observation summaries and vitals. This would demonstrate how to correlate published research with real-world clinical data, potentially uncovering interesting links, trends or discrepancies. Might be helpful for care-givers or nutrition coaches to link research data to their cases.

Tapping into existing experiment logs:

By referencing specific log IDs, the agent could reference current insights to past experimental results, providing added context from the local docs.

Which use-case do you think has the most potential to showcase the mcp integration?

[Croskcool]

Hook mcp servers for contextual information that is not available on the web.

[rlancemartin]

100%. We need to do this. @bartolli mentioned he may put up a PR.

Also chatted with @vbarda who worked on MCP tools with LangGraph.

I will also have a look in the coming days.

[rlancemartin]

This should be more feasible now that I added an agent implementation, which uses tools.

[bartolli]

@rlancemartin The PR is ready for review.

LangSmith thread check: https://smith.langchain.com/public/2922d8ab-b773-468e-a2af-80b7dbd2f4ab/r

PR: https://github.com/langchain-ai/open_deep_research/pull/95

[rlancemartin]

Added to multi_agent.py.