brief-mcp

Know everyone in the room before you walk in.

brief-mcp is the first open-source Model Context Protocol server for meeting intelligence. It connects Claude — and any MCP-compatible AI assistant — to your calendar, the web, your CRM, and your own meeting history, then generates executive-grade briefing cards in seconds.

No browser tabs. No LinkedIn spelunking. No manual prep.

Table of Contents

• What it does

• Demo

• Why brief-mcp is different

• Install

• Quick Start

• Client Configuration

  • Claude Desktop

  • VS Code / Cursor

  • Zed

• Integrations

  • Google Calendar

  • Microsoft 365 / Outlook

  • HubSpot CRM

  • Web Research

  • Meeting Memory

• All 13 Tools

  • Pre-Meeting: Google Calendar

  • Pre-Meeting: Microsoft 365

  • CRM Enrichment

  • Post-Meeting Memory

• Usage Flows

• Architecture

• Privacy

• Contributing

• Roadmap

• License

Related MCP server: Google MCP Server

What it does

brief-mcp gives your AI assistant five capabilities it doesn't have out of the box:

Demo

You open Claude and type: "Brief me for my 2pm call."

Claude responds:

============================================================
  MEETING BRIEF: Q4 PARTNERSHIP REVIEW
============================================================

  When    : Tuesday, October 13 2026 at 02:00 PM
  Duration: 60 minutes
  Link    : https://meet.google.com/abc-defg-hij

  Attendees (3): Alice Chen (you), Bob Patel, Carol Wu

------------------------------------------------------------
  ATTENDEE PROFILES
------------------------------------------------------------

  >> Bob Patel
     Role    : VP of Engineering @ PartnerCo
     Bio     : Engineering leader, 15 years in distributed systems.
               Previously Staff Engineer at Google and DataStream Inc.
     Recent  :
       - Bob Patel promoted to VP of Engineering (Sept 2026)
       - PartnerCo Engineering Blog: new latency benchmarks published
     Co News :
       - PartnerCo raises $50M Series B to accelerate AI platform
       - PartnerCo announces APAC expansion, opens Singapore office

  >> Carol Wu
     Role    : Director of Product @ PartnerCo
     Bio     : Product leader, ex-Google PM for Firebase.
     Recent  :
       - Carol Wu joins PartnerCo from Google as Director of Product
       - Carol Wu keynote at ProductCon SF 2026

------------------------------------------------------------
  SUGGESTED TALKING POINTS
------------------------------------------------------------
  1. Ask Bob about engineering priorities and how the integration fits
     their new Kubernetes-native platform.
  2. Reference the Series B — congratulate them and probe roadmap changes.
  3. Carol is new from Google — don't assume prior integration context.
  4. Confirm next-steps owner before the call ends.

------------------------------------------------------------
  WATCH-OUTS / SENSITIVITIES
------------------------------------------------------------
  ! Series B funding may prompt renegotiation — know your commercial floor.
  ! Carol joined in 2026 — bring her up to speed on integration history.

  Generated: 2026-10-13 09:14
============================================================

Or for a returning customer meeting:

------------------------------------------------------------
  PAST INTERACTION HISTORY — Bob Patel
------------------------------------------------------------
  * Met 2026-07-18 — Product roadmap sync
    Committed: "Send updated API latency benchmarks by Aug 1"
    Status: Follow-up sent ✓

  * Met 2026-09-02 — Q3 review
    Committed: "Schedule infra deep-dive with their SRE team"
    Status: OPEN — 41 days outstanding

  Open action items from past meetings: 1
------------------------------------------------------------

All of that from a single prompt. No switching tools, no browser tabs.

Why brief-mcp is different

The only meeting intelligence MCP server

As of October 2026, brief-mcp is the only open-source MCP server dedicated to meeting preparation, CRM enrichment, and post-meeting memory. A GitHub-wide search returns zero competing implementations.

End-to-end meeting lifecycle

Most tools address one phase. brief-mcp covers all three:

Before  →  During context  →  After
─────────────────────────────────────────────────────────
Brief      Who's in the room,   Debrief, action items,
calendar   news, CRM context,   follow-up email, memory,
           past history         open item tracking

Dual-calendar: Google + Microsoft 365

Supports both ecosystems with separate OAuth2 flows — Google Calendar via the Workspace API and Outlook via Microsoft Graph (MSAL device-code, headless-friendly). Most professionals use one or the other; brief-mcp supports both from one server.

Persistent relationship memory

brief-mcp builds a local SQLite database (with FTS5 full-text search) of every meeting you record. Before a repeat meeting, brief_with_history surfaces:

• Every past meeting with that person

• Commitments made and whether they were fulfilled

• Topics that recur across sessions

• How many days outstanding open action items have been sitting

No SaaS subscription. No cloud sync. Fully yours.

Engineered for correctness

• Parallel async research — asyncio.gather fires all attendee lookups simultaneously; a 6-person meeting takes the same wall-clock time as a 1-person meeting

• Timezone-correct everywhere — Microsoft Graph timezone strings resolved via ZoneInfo IANA and Windows→IANA mapping; all datetimes are timezone-aware throughout

• TTL cache — 60-minute, 256-entry cache with dual-indexing (name+company and name-only) prevents cache misses when calendar lookups don't have company context

• Full pagination — Google Calendar and Microsoft Graph results are paginated; no silent truncation at 25 items

• 123 tests, all offline — every integration is mocked; no credentials needed to run the full suite

Install

Zero-install (recommended)

uvx brief-mcp

uvx runs brief-mcp in an isolated environment without a permanent install. Requires uv.

pip

pip install brief-mcp

From source

git clone https://github.com/MLDreamer/brief-mcp
cd brief-mcp
uv sync
uv run brief-mcp

Quick Start

Step 1 — Enable Google Calendar (5 minutes, one-time):

Go to Google Cloud Console → APIs & Services → Enable Google Calendar API → Credentials → Create OAuth 2.0 Client ID (Desktop app) → Download JSON.

Step 2 — Create .env in your working directory:

# Required for Google Calendar
GOOGLE_CREDENTIALS_PATH=/path/to/credentials.json

# Optional but recommended — free tier at brave.com/search/api
BRAVE_API_KEY=BSAxxx...

# Optional — Microsoft 365 / Outlook
AZURE_CLIENT_ID=your-azure-app-client-id
AZURE_TENANT_ID=common

# Optional — HubSpot CRM
HUBSPOT_API_KEY=pat-na1-xxxx...

Step 3 — Configure your AI client (see section below).

Step 4 — Open Claude and say:

"Brief me for today's meetings."

On first run, a browser window opens for Google OAuth consent. After that, fully silent.

Client Configuration

Claude Desktop

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "brief-mcp": {
      "command": "uvx",
      "args": ["brief-mcp"],
      "env": {
        "GOOGLE_CREDENTIALS_PATH": "/Users/you/credentials.json",
        "BRAVE_API_KEY": "BSAxxx...",
        "AZURE_CLIENT_ID": "your-client-id",
        "HUBSPOT_API_KEY": "pat-na1-xxxx..."
      }
    }
  }
}

Restart Claude Desktop after saving.

VS Code / Cursor

Create .vscode/mcp.json in your workspace root:

{
  "servers": {
    "brief-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": ["brief-mcp"],
      "env": {
        "GOOGLE_CREDENTIALS_PATH": "${workspaceFolder}/credentials.json",
        "BRAVE_API_KEY": "BSAxxx...",
        "HUBSPOT_API_KEY": "pat-na1-xxxx..."
      }
    }
  }
}

Zed

In ~/.config/zed/settings.json:

{
  "assistant": {
    "mcp_servers": {
      "brief-mcp": {
        "command": "uvx",
        "args": ["brief-mcp"],
        "env": {
          "GOOGLE_CREDENTIALS_PATH": "/path/to/credentials.json",
          "BRAVE_API_KEY": "BSAxxx..."
        }
      }
    }
  }
}

Integrations

Google Calendar

Auth: OAuth2 desktop flow via google-auth-oauthlib. Token stored at ~/.brief-mcp/token.json and refreshed silently.

Required env var: GOOGLE_CREDENTIALS_PATH — path to the JSON file downloaded from Google Cloud Console.

Scopes used: calendar.readonly (optionally gmail.readonly for prior-email context).

Graceful degradation: If not configured, all Google Calendar tools return a clear message and continue working — research_attendee functions without any calendar.

Microsoft 365 / Outlook

Auth: MSAL device-code flow — headless-friendly, works in any terminal. Token stored at ~/.brief-mcp/ms_token.json and auto-refreshed.

Required env vars:

AZURE_CLIENT_ID=your-app-client-id
AZURE_TENANT_ID=common          # "common" = personal + work accounts

Azure Portal setup (one-time, 3 minutes):

1. Azure Portal → App registrations → New registration

2. Supported account types: Accounts in any organizational directory and personal Microsoft accounts

3. Redirect URI → Platform: Mobile and desktop → https://login.microsoftonline.com/common/oauth2/nativeclient

4. API Permissions → Microsoft Graph → Delegated → add Calendars.Read + offline_access → Grant admin consent

5. Copy Application (client) ID → set as AZURE_CLIENT_ID

On first run, brief-mcp prints:

To sign in, visit https://microsoft.com/devicelogin and enter code: ABCD1234

After sign-in, all subsequent runs are silent.

Timezone handling: Microsoft Graph returns naive datetime strings with a separate timeZone field. brief-mcp resolves this using Python's zoneinfo for IANA names and a built-in Windows→IANA mapping for timezones like "Eastern Standard Time".

HubSpot CRM

Auth: Private app token (no OAuth callback required).

Required env var: HUBSPOT_API_KEY

HubSpot setup (5 minutes):

1. HubSpot → Settings → Integrations → Private Apps → Create a private app

2. Scopes: crm.objects.contacts.read + crm.objects.deals.read

3. Generate token → copy → set as HUBSPOT_API_KEY

What it fetches: Contact name, email, job title, company, lifecycle stage, last contacted date, and all associated deals (name, stage, amount, close date).

Graceful degradation: If HUBSPOT_API_KEY is not set, CRM tools return a helpful setup message instead of crashing.

Web Research

Primary: Brave Search API — set BRAVE_API_KEY for best results (free tier available, 2,000 queries/month).

Fallback: DuckDuckGo HTML endpoint — works with no configuration, no API key required. brief-mcp automatically falls back if Brave is not configured.

What it does per person: Fires 3 parallel searches — general bio, LinkedIn profile, and company news — then extracts current role, company, recent news headlines, and background from the combined results.

Meeting Memory

No configuration required. Memory is stored automatically at ~/.brief-mcp/memory.db — a local SQLite database using FTS5 full-text search.

The database stores:

• Meeting records (title, date, attendees, summary, action items, outcomes)

• Per-person interaction history (topics discussed, sentiment, commitments)

• Action item status with per-item completion tracking

Data never leaves your machine.

All 13 Tools

Pre-Meeting: Google Calendar

get_upcoming_meetings

Reads your Google Calendar and returns all meetings in the next N hours.

Arguments:
  hours_ahead: int = 24     # Look-ahead window in hours (max: 168 / 1 week)

Returns:
  List of meetings, each with:
  - id, title, start_time, end_time (timezone-aware)
  - attendees: [{name, email, is_organizer}]
  - meeting_url (Google Meet, Zoom, or Teams link if present)
  - location, description, calendar_link

research_attendee

Researches a single person via parallel web search. Results are cached for 60 minutes.

Arguments:
  name: str           # Full name (required)
  company: str = ""   # Company name — improves search accuracy significantly
  context: str = ""   # Any extra context you know about this person

Returns:
  PersonProfile with:
  - current_role, company, background, linkedin_summary
  - recent_news[]    — recent headlines about this person
  - company_news[]   — recent headlines about their company
  - research_sources[]

get_meeting_brief

Looks up a meeting from your calendar and generates a full brief — attendee profiles, talking points, and watch-outs — all researched in parallel.

Arguments:
  meeting_title: str = ""   # Partial title match (case-insensitive)
  meeting_id: str = ""      # Exact ID from get_upcoming_meetings (takes priority)
  hours_ahead: int = 24

Returns:
  Formatted brief card as a multi-line string.
  Returns an error with available meeting titles if no match found.

brief_all_today

Generates briefs for every meeting on today's calendar. Deduplicates attendee research across meetings — if the same person appears in your 10am and 3pm, they are researched exactly once.

Arguments: none

Returns:
  Full daily briefing — one card per meeting, separated by dividers.
  Header shows total meeting count and date.

brief_with_history

Same as get_meeting_brief but enriched with past interaction history from your local meeting memory. Shows what was discussed, committed, and decided in every prior meeting with each attendee.

Arguments:
  meeting_title: str = ""
  meeting_id: str = ""
  hours_ahead: int = 24

Returns:
  Standard brief card + past interaction section per attendee.
  Shows open action items, recurring topics, and commitment status.

Pre-Meeting: Microsoft 365

get_outlook_meetings

Reads your Outlook / Microsoft 365 calendar via Microsoft Graph. Returns the same format as get_upcoming_meetings.

Arguments:
  hours_ahead: int = 24

Returns:
  Same format as get_upcoming_meetings.
  Requires AZURE_CLIENT_ID in env.

get_outlook_brief

Generates a full brief from your Outlook calendar. Identical output format to get_meeting_brief.

Arguments:
  meeting_title: str = ""
  meeting_id: str = ""
  hours_ahead: int = 24

Returns:
  Formatted brief card string.

CRM Enrichment

get_crm_context

Looks up a person in HubSpot CRM and returns their contact record and associated deals.

Arguments:
  name: str           # Full name
  email: str = ""     # Email address — used for exact-match lookup if provided
  company: str = ""   # Company name — improves search when name is common

Returns:
  Formatted CRM summary:
  - Name, job title, company, lifecycle stage
  - Last contacted date
  - All associated deals: name, stage, amount, close date
  Returns "No CRM record found for {name}" if not found.

enrich_brief_with_crm

Generates a standard meeting brief then appends a CRM section for every attendee found in HubSpot. Ideal for sales and partner meetings.

Arguments:
  meeting_title: str   # Matches against Google Calendar

Returns:
  Full brief card + CRM CONTEXT section at the bottom.
  Shows deal pipeline, stage, amount, and last activity per attendee.

Post-Meeting Memory

debrief_meeting

Records what happened in a meeting. Saves to the local SQLite database, auto-extracts action items from free-form notes, and drafts a follow-up email.

Arguments:
  meeting_title: str      # Matches the calendar event (looks back 24h + ahead 2h)
  notes: str              # Raw meeting notes — bullet points, transcript, anything
  action_items: str = ""  # Comma-separated tasks (e.g. "Alice to send report, Bob to schedule demo")
  outcomes: str = ""      # What was decided or agreed

Returns:
  Formatted debrief card:
  - Structured meeting record saved to memory.db
  - Extracted action items with inferred owners and due dates
  - Draft follow-up email ready to send
  - Next-step suggestions

Action item extraction recognises:
  "ACTION: ..."  "TODO: ..."  "FOLLOW-UP: ..."
  "- [ ] ..."    "* [ ] ..."

search_meeting_history

Full-text search across all past meetings using SQLite FTS5. Searches meeting title, attendees, summary, action items, and outcomes simultaneously.

Arguments:
  query: str        # Keyword, person name, company, topic, or date
  limit: int = 10

Returns:
  Formatted list of matching meetings with date, attendees, summary, and action items.
  Returns "No matching meetings found" if empty.

get_person_history

Returns the complete interaction history with a specific person — every meeting, every topic, every commitment, in chronological order.

Arguments:
  name: str         # Person's name (partial match supported)

Returns:
  Chronological timeline of all meetings with this person:
  - Meeting date, title, topics discussed
  - Commitments made (by you and by them)
  - Action item status (open / closed / how many days outstanding)

get_open_action_items

Lists every open action item across all past meetings, with owner, meeting context, and how long it has been outstanding.

Arguments: none

Returns:
  Sorted list of open action items:
  - Task description
  - Owner (if detected from notes)
  - Source meeting title and date
  - Days since the meeting
  Returns "No open action items" if all tasks are complete.

Usage Flows

Morning briefing

"Give me a morning brief for all my meetings today."

brief-mcp calls brief_all_today() — reads your calendar, researches every attendee in parallel (each person once, even across overlapping meetings), and returns one formatted card per meeting. Typical time for a 5-meeting day with 12 unique attendees: under 8 seconds.

Before a specific meeting

"Brief me for the Q4 review with PartnerCo at 2pm."

Use brief_with_history for recurring meetings — it adds a full history section showing everything said and committed in prior sessions with those same people.

After a meeting

"Debrief the Q4 review. Notes: discussed roadmap integration, 
Bob will send technical spec by Oct 20, Carol to set up infra call.
Outcomes: agreed on Phase 2 timeline."

brief-mcp:

• Saves the meeting record to ~/.brief-mcp/memory.db

• Extracts action items: Bob → send technical spec by Oct 20, Carol → set up infra call

• Drafts a follow-up email with the agreed outcomes and next steps

• Next time you meet Bob or Carol, brief_with_history will surface this record

Sales meeting with CRM context

"Enrich the brief for my Salesforce call with deal context from HubSpot."

enrich_brief_with_crm adds a CRM section showing the deal name, stage, amount, and last activity date for every attendee found in HubSpot — giving you the commercial picture alongside the personal one.

Ad-hoc person research

"Who is Priya Mehta from Infosys?"

research_attendee("Priya Mehta", "Infosys") — runs in under 3 seconds, cached for 60 minutes.

Checking outstanding commitments

"What's still open from my meetings this week?"

get_open_action_items() scans the memory database and returns every unresolved task with owner, source meeting, and how long it has been outstanding.

Architecture

brief-mcp/
├── server.py                   FastMCP server — registers all 13 tools
├── models.py                   Pydantic v2 models (Meeting, PersonProfile, MeetingBrief)
├── cache.py                    TTL cache — 60 min / 256 entries, dual-indexed
├── memory.py                   SQLite + FTS5 meeting memory database
│
├── integrations/
│   ├── google_calendar.py      Google Calendar OAuth2 client
│   ├── outlook_calendar.py     Microsoft Graph MSAL client + timezone resolution
│   ├── web_search.py           Brave Search + DuckDuckGo fallback (3 parallel queries)
│   ├── hubspot.py              HubSpot CRM REST client
│   └── gmail.py                Optional Gmail prior-email context
│
└── tools/
    ├── _shared.py              Shared helpers (_find_meeting, _research_attendee_with_cache)
    ├── calendar_tool.py        get_upcoming_meetings
    ├── research_tool.py        research_attendee
    ├── brief_tool.py           get_meeting_brief + brief_all_today + helpers
    ├── outlook_tool.py         get_outlook_meetings + get_outlook_brief
    ├── crm_tool.py             get_crm_context + enrich_brief_with_crm
    └── debrief_tool.py         debrief_meeting + memory tools (5 tools)

Key design decisions

Parallel async research. asyncio.gather fires all attendee lookups simultaneously. A meeting with 6 attendees takes the same wall-clock time as a meeting with 1. brief_all_today deduplicates across meetings first, then fires a single gather across all unique attendees for the day.

Dual-indexed TTL cache. Profiles are stored under both name+company and name-only keys. When the calendar fires lookups without company context (the common case), it still hits profiles stored by the user with explicit company. 60-minute TTL, 256-entry LRU.

Timezone-correct datetimes. Google Calendar returns ISO 8601 strings (timezone-aware). Microsoft Graph returns naive strings with a separate timeZone field (which may be an IANA name or a Windows display name). brief-mcp resolves both via zoneinfo.ZoneInfo and a built-in Windows→IANA mapping. All internal datetimes are timezone.utc-aware.

FTS5 full-text search. Meeting memory uses SQLite's built-in FTS5 extension — no external search dependency. The UPDATE trigger is implemented as DELETE + INSERT (FTS5 does not support UPDATE statements). Search covers title, attendees, summary, action items, and outcomes simultaneously.

Per-item action tracking. mark_action_done supports both item-level completion (stores completed item text in a JSON column) and meeting-level closure (follow_up_sent). get_open_action_items filters out completed items per meeting.

Graceful degradation everywhere. Every integration checks for its required env var at call time. Missing credentials return a clear setup message — the server never crashes, and unconfigured integrations don't affect configured ones.

Privacy

All processing is local. Your meetings, notes, and history never leave your machine.

No telemetry. No analytics. No cloud sync. No accounts.

Contributing

Pull requests are welcome. Please open an issue to discuss significant changes before starting.

Development setup

git clone https://github.com/MLDreamer/brief-mcp
cd brief-mcp
uv sync --dev

Running tests

All 123 tests run offline — no credentials required.

uv run pytest                          # full suite
uv run pytest tests/test_memory.py    # specific module
uv run pytest -v --tb=short           # verbose output

Linting

uv run ruff check src/ tests/
uv run ruff format src/ tests/

Test coverage

Roadmap

Shipped

• Google Calendar — OAuth2, full pagination, Meet/Zoom/Teams URL extraction

• Microsoft 365 / Outlook — MSAL device-code, Windows timezone mapping, pagination

• Web research — Brave Search + DuckDuckGo fallback, 3 parallel queries per person

• Parallel async research with asyncio.gather + TTL cache

• HubSpot CRM — contact search, deal pipeline, lifecycle stage

• brief_all_today with cross-meeting attendee deduplication

• Post-meeting debrief — action item extraction, follow-up email draft

• SQLite FTS5 meeting memory — search, person history, action item tracking

• brief_with_history — pre-meeting brief + full past interaction context

• Optional Gmail prior-email context enrichment

• 123 tests (all offline, all passing)

Planned

• Salesforce CRM connector

• Slack — post brief or debrief to a channel

• LinkedIn data via official API (when available)

• Automatic follow-up email sending (Gmail / Outlook)

• Voice brief via text-to-speech

• Team shared memory (multi-user SQLite / shared drive)

• Apollo.io / People Data Labs for enriched professional profiles

Star History

Similar Projects

• blender-mcp — AI controls Blender 3D

• codebase-memory-mcp — AI-powered code intelligence

License

MIT — see LICENSE.

Built on FastMCP and the Model Context Protocol.

If brief-mcp saves you time before a meeting, consider starring the repo — it helps others find it.