evolutiondb-teams-sync 0.1.0

Sync Microsoft Teams 1:1 + group chat messages into EvolutionDB long-term memory.

evolutiondb-teams-sync

Stream Microsoft Teams 1:1 and group chat messages into EvolutionDB so they become part of the long-term memory the MCP server already exposes to Claude, ChatGPT, Gemini, and Copilot. After running the sync, "what did Ahmet say last week about the release plan" becomes a normal search_memory call from any MCP-aware client.

Channel messages, meeting transcripts, and call recordings are out of scope. They need higher-privilege Graph permissions that require tenant admin consent — this client deliberately stays inside scopes a user can self-consent to.

---

Before you enable this

A few non-technical things to settle first:

• Other people's words land in the memory too. A 1:1 chat or group chat captures everyone's messages, not just yours. Some workplaces treat external archiving of colleagues' messages as a policy violation independent of the technical access controls. This is an HR / legal question, not a permissions question.
• Microsoft sees the request. Every Graph API call is logged against your account in the tenant audit log. Your IT team can see that an app is reading your chats.
• Token revocation is one click. You can pull the plug at any time from https://myaccount.microsoft.com/applications — the sync daemon's next call gets a 401 and stops.

If those are fine in your environment, continue.

---

One-time setup

1. Register an Entra ID app

1. Open https://portal.azure.com → Microsoft Entra ID → App registrations → New registration. (Microsoft renamed Azure Active Directory to Microsoft Entra ID in 2023; legacy docs that say "Azure AD" mean the same thing.)
2. Name: EvolutionDB Memory Sync (or anything you like).
3. Supported account types: Accounts in this organizational directory only (single tenant) is enough.
4. Redirect URI: leave blank — the device-code flow doesn't use one.
5. After registration, open Authentication → enable Allow public client flows → save.
6. Open API permissions → Add a permission → Microsoft Graph → Delegated permissions → tick:
   • Chat.Read
   • User.Read
   • offline_access
7. Copy the Application (client) ID and the Directory (tenant) ID from the app's Overview page.

None of these scopes require tenant admin consent — you can grant them to yourself the first time you log in.

2. Configure the sync

cd client/teams-sync
cp .env.example .env
$EDITOR .env   # paste AZURE_TENANT_ID and AZURE_CLIENT_ID

3. Install the package

From PyPI (recommended once a release is up there):

pip install evolutiondb-teams-sync

Or from source if you're working on the code:

pip install -e .

4. Log in once

python -m teams_sync.sync --auth

The CLI prints a Microsoft login URL and a short code. Open the URL, paste the code, sign in with your work account, approve the permissions. The refresh token lands in ~/.evosql/teams_token_cache.json (mode 0600).

---

Daily use

# One-shot, last 24 hours (uses the per-chat watermark on later runs):
python -m teams_sync.sync --once

# Daemon, polls every 10 minutes:
python -m teams_sync.sync --interval 600

# Verify auth + scopes without writing to the DB:
python -m teams_sync.sync --once --dry-run

--since 1h | 24h | 7d | 30m controls how far back the first run on a chat goes. After that, the per-chat watermark in EvolutionDB takes over, so reruns never re-ingest the same message.

---

Running in Docker

The repo's demo compose file ships a teams-sync service behind a profile so it doesn't start with the rest of the demo:

# First-run device-code login (runs once, cleans up after itself):
docker compose -f docker-compose.demo.yml --profile teams \
    run --rm teams-sync --auth

# Bring the daemon up:
docker compose -f docker-compose.demo.yml --profile teams up -d teams-sync

The token cache lives on a named volume (teams_sync_state), so container recreates don't force you to log in again.

---

How a synced message looks in memory

Each Teams message becomes one memory row in the mcp_mem store (or whatever TEAMS_MEMORY_STORE points to):

{
  "fact":         "Ahmet Yılmaz: Stagingde regression testi başarısız",
  "source":       "teams",
  "kind":         "chat",
  "chat_id":      "19:meeting_…",
  "chat_name":    "Backend standup grubu",
  "chat_type":    "group",
  "message_id":   "1715064862123",
  "sender":       "Ahmet Yılmaz",
  "sender_id":    "<aad-object-id>",
  "text":         "Stagingde regression testi başarısız",
  "created_at":   "2026-05-07T09:14:22Z",
  "modified_at":  "2026-05-07T09:14:22Z",
  "tags":         ["teams", "chat", "Backend standup grubu"]
}

Per-chat watermarks live alongside the messages, keyed teams_state_<chat_id>. The store wrapper hides them from message counters and they're filtered out by the MCP search_memory tool because their fact field is empty.