Beever Atlas MCP
MCP server for the knowledge base Beever Atlas builds from Slack, Discord, Teams and Mattermost chats: cited answers, channel wikis, a graph
- 451 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Reads work conversations in full, which can hold personal data and trade secrets
- Stores Slack, Discord and Teams tokens, and messages go to the chosen LLM provider for fact extraction
- The trigger_sync tool starts channel syncs, and an MCP key opens the knowledge base to anyone who holds it
Install
- Works in
Manual install
git clone https://github.com/beever-ai/beever-atlas.git && cd beever-atlas && ./atlasThe installer asks for keys and providers, generates secrets and starts the stack through Docker Compose. The dashboard opens at localhost:3000.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Reads work conversations in full, which can hold personal data and trade secrets
- Stores Slack, Discord and Teams tokens, and messages go to the chosen LLM provider for fact extraction
- The trigger_sync tool starts channel syncs, and an MCP key opens the knowledge base to anyone who holds it
This is third-party code. Review the repository files before installing.
What it does
Beever Atlas pulls a team's conversations from Slack, Discord, Microsoft Teams and Mattermost, extracts atomic facts, removes duplicates and clusters them into topic pages that cite the source messages. A Neo4j graph links the people, decisions and projects mentioned across channels. Each channel gets a wiki that updates as new messages arrive, and natural-language questions get answers with citations. For agents, Atlas exposes an MCP server at /mcp so Claude Code, Cursor and other clients can query the knowledge base without the dashboard. The catalog holds 28 tools: discovery of connections and channels (whoami, list_connections, list_channels), search and questions (ask_channel with quick, deep and summarize modes, search_memory, search_channel_facts), reading wiki pages and the graph, plus channel sync and conversation sessions. The stack runs through Docker Compose: a backend, a bot and a web dashboard on port 3000, backed by Weaviate, Neo4j, MongoDB and Redis. There is also a light mode: a stdio build of the MCP server without datastores serves the tool catalog but cannot read knowledge with no database behind it.
Who it is for. For teams whose decisions and agreements sink into chat history, and for developers who want Claude Code or Cursor to read that memory.
Good fit when
- A new teammate needs a short overview of a Slack or Discord channel instead of three months of history
- An agent in Cursor has to answer how auth works here, based on the team's discussions
- You need to find who made a decision and when, with a link to the original message
- Chats from several platforms have to land in one knowledge base
- You want to inspect the MCP tool set without deploying the whole stack: the stdio mode does that
Not a fit when
- The team will not send conversations to outside models: fact extraction goes through your chosen LLM provider
- You want one small MCP server with no databases: Atlas starts Weaviate, Neo4j, MongoDB and Redis
- Your chats live in messengers outside Slack, Discord, Teams and Mattermost
- You need MCP without TLS: in production the /mcp endpoint must be served over HTTPS
Example request
Ask Atlas about the devops channel why we moved to the new load balancer and show the quoted messagesMCP
- Transport
- http, stdio
- Authentication
- API key
Environment variables
BEEVER_MCP_ENABLEDrequired- true turns on the /mcp mount, off by default
BEEVER_MCP_API_KEYSrequired, secret- Comma-separated keys for external agents, one per agent instance
GOOGLE_API_KEYrequired, secret- Gemini key for fact extraction and answers
JINA_API_KEYrequired, secret- Jina key for semantic search embeddings
Limitations
Installation needs Docker with Compose and keys for two providers: GOOGLE_API_KEY for Gemini and JINA_API_KEY for embeddings. The ./atlas installer also lets you pick other providers, local Ollama included.
- Slack, Discord and Teams tokens are set in the web dashboard and stored encrypted in MongoDB.
- MCP keys (BEEVER_MCP_API_KEYS) must differ from the API and bot keys or the server refuses to start. Rotating a key needs a restart.
- Defaults in .env.example suit a local trial: before a team deployment replace the keys and set BEEVER_ENV=production.
- The /mcp endpoint is off by default; BEEVER_MCP_ENABLED=true turns it on.
How to disable. Remove the beever-atlas entry from the client's MCP settings and turn off BEEVER_MCP_ENABLED. docker compose down stops the stack, and docker compose down -v also deletes all indexed data.
FAQ
How do I connect Beever Atlas to Claude Code?
Set BEEVER_MCP_ENABLED=true, define BEEVER_MCP_API_KEYS and add a server to .mcp.json with a url like https://atlas.example.com/mcp, transport streamable-http and an Authorization: Bearer header with your key.
How does the stdio mode differ from /mcp?
python -m beever_atlas.api.mcp_server (or beever-atlas-mcp) serves the same tool catalog over stdio with no HTTP and no datastores. The catalog answers instantly, while calls that read knowledge return a structured error without databases.
What data does Atlas take from chats?
Messages from the channels you sync: it builds facts, topics, entities and relationships from them. The user chooses which channels to sync, and MCP answers are filtered by the key's access to connections.
What do I need to run Atlas locally?
Docker with Compose and two keys, GOOGLE_API_KEY and JINA_API_KEY. make demo starts the stack with a ready demo corpus and needs no keys for seeding, while questions through /api/ask need a Gemini key.
How many tools does the MCP server have?
Twenty-eight: three for discovery, seventeen for search and reading, three for the graph, one for sessions and three for long operations. The suggested order is whoami, list_connections, list_channels, then the rest.
Discussion
No questions or reviews yet.
Related
MCP serversThe ATLAS MCP server keeps an agent's projects, tasks and knowledge in a Neo4j graph database: dependencies, search and a deep research plan
Official Anytype MCP server: an agent searches, creates and edits objects, lists and types in your knowledge base
WEEEK MCP with the knowledge base
weeek-mcp
A WEEEK MCP server: tasks via the public API and full knowledge-base CRUD via the internal API with a saved browser session
Notion MCP (awkoy)
Notion MCP Server (awkoy)
Notion MCP on a token: an agent searches pages, runs databases, edits blocks and comments through three tools and 47 operations, with no manual sign-in
Atlassian's official cloud MCP server for Jira, Confluence, Jira Service Management, Bitbucket, Compass and Loom, with skills
An MCP server for Slack: channel history, threads, DMs, search and unreads, with posting off by default
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Beever Atlas (this entry) | yes | API key | 451 | |
| LLM Wiki | no | 1,473 | a plugin that compiles a wiki from notes and sources you import yourself, with no chat sync | |
| Deepwiki MCP | no | not required | 1,386 | fetches ready pages from deepwiki.com about public repositories instead of your chats |
README in short
The page shows how Atlas turns chats into a wiki: six short demos (connecting platforms, sync, memory, wiki, questions, MCP), a diagram with two memory stores and the reasoning behind wiki-first retrieval. Three install routes follow (the ./atlas installer, manual Docker, local development), the required keys and a table of steps before production. A separate section covers the MCP server for external agents and points to docs/mcp-server.md. The license is Apache 2.0.