The MCP server — your AI agent has meeting context now

If you're already running Claude Desktop or an MCP-aware client, here's the short version: drop `https://mcp.briefi.sh/mcp` into your config, paste your per-tenant Briefi API key as a Bearer token, and your agent has eleven tools. List meetings. Get a meeting. Generate a brief. Generate a follow-up. List contacts. List deals. Draft a reply. Push to CRM. Done.

The longer version — why this matters for operators, not just for engineers — is below.

Context is the unlock

An AI agent without context is a clever chatbot. The same agent with read access to your meetings, contacts, deals, and recent threads is the operator's dream: 'Summarise my Northwind relationship over the last three calls and draft a follow-up that addresses the pricing concern.' That used to require fifteen minutes of context-stuffing — copy this transcript, paste that email, attach the deal stage. With the Briefi MCP server it's one prompt.

The agent isn't doing magic. It's doing what you would do, faster, because the substrate is already there. The same is true of 'Find every meeting from last week where the customer mentioned pricing and tag them for a discount conversation.' Either you spend half an hour with grep and a spreadsheet, or your agent runs the query in three seconds.

Dull on purpose

The implementation is dull on purpose. The MCP function proxies to the same v1 REST surface that powers our public API — same auth, same rate limits, same RLS. There's no admin-client shortcut, no special path for the AI. Your agent reads what you read, gated by the same per-tenant API key, capped at one hundred requests per minute per workspace.

Dull also means safe. The AI can't see workspaces you're not a member of. It can't write to objects you don't own. It can't bypass plan gates. If you revoke the API key, the agent loses access immediately. The same controls that protect your team-mates from each other protect you from your agent.

Two transports, one tool list

The hosted server at `mcp.briefi.sh/mcp` uses Streamable HTTP — works inside Claude Desktop, claude.ai web, and any modern MCP client. No installation needed, no local process to manage. The local stdio server (`npx @briefi/mcp-server`) is for older clients that don't yet speak HTTP. Both ship the same eleven tools, mapped to the same v1 endpoints. Pick the transport your client supports and you're done.

What you can actually do with it

The interesting use cases aren't the obvious ones. 'Generate a brief' is fine — you can do that in the UI. The valuable patterns are the ones the UI doesn't surface: agent-driven multi-step workflows that cross meetings, contacts, and deals in a single prompt.

Some examples I use myself: 'Pull every meeting from the last quarter where we discussed integrations, draft a follow-up that proposes a quarterly check-in.' 'Find all the deals stuck at the qualification stage for more than fourteen days and write a re-engagement email per stalled contact.' 'Look at next week's calendar, identify the three highest-value meetings, and walk me through the prep for each.'

Why we shipped this

We didn't ship this because everyone else has one — almost nobody in the meeting-AI category does. We shipped it because I wanted my own Claude to read my own meetings without me copy-pasting briefs into the chat. Now it does, and we've made it available to anyone on the Business plan.

Available now to Business-plan workspaces. Setup takes thirty seconds — see the integrations page for the connector list, or pricing for which plan unlocks the API. If you missed why I built Briefi in the first place, that's the context for why a meeting-AI tool ships its own MCP server.