Granola Adds Search_meetings to Its MCP.

Granola’s new search_meetings tool helps AI agents find relevant meetings and retrieve details from their transcripts. It replaces a date-only discovery process that often left agents working from a small sample of meeting notes.

Previously, an agent connected through Granola’s Model Context Protocol could use list_meetings to filter meetings by date. It then had to open meetings individually and pull each full transcript into its context window.

Faced with 30 meeting notes, agents would sometimes answer after reading six or seven. Granola users worked around this by manually copying transcripts into their chosen agent.

Granola explains the rebuild in its engineering blog post. The work focused on two tasks: finding the right meeting, then finding the relevant statements inside it.

search_meetings can return meetings in date order when the agent supplies no keywords. When it receives a query, it runs BM25 search against an index stored in Turbopuffer. It supports keyword and people searches that were already available in Granola’s in-app chat but had not been brought to its MCP.

This gives connected agents a narrower set of meetings to inspect before producing an answer. It also makes transcript details available for questions where the final note does not contain the full discussion, such as numerical pricing points raised during a customer call.

The update complements Granola Agent Briefs Give AI Agents Daily Work Context, which packages recent meeting context into a daily briefing. search_meetings handles targeted retrieval when an agent needs to investigate a specific topic or person.

Granola found that the same question could produce different answers across clients including Claude Desktop and Codex. It built an MCP evaluation harness to turn those differences into measurable tests.

The harness checks whether an agent retrieves a required fact, retains that fact in its working context and uses it accurately in the final answer. Granola used ROUGE-L recall and LLM judges to score those stages.

This testing exposed a second problem. Providing a transcript index did not guarantee that an agent would use it. Granola also worked on its tool descriptions so agents would choose transcript retrieval when a question depended on what participants actually said, rather than relying only on the meeting note.

What is Granola search_meetings?

search_meetings is an MCP tool that lets AI agents find Granola meetings by keyword, person or date. Keyword queries use BM25 search over an index stored in Turbopuffer.

How does search_meetings compare with list_meetings?

list_meetings filtered meetings by date. search_meetings adds keyword and people search, while still returning a date-ordered list when no keywords are supplied.

Is Granola search_meetings free?

Granola’s post does not state which plans include search_meetings or provide separate pricing for it.