M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
B

Beever Atlas

Community
Open-source LLM knowledge base that turns team chats into a typed knowledge graph and auto-generated wiki.
How to install ↓ Compare GitHub source repository ↗
Category
Other
Stars
★ 449 Very popular
Transport
stdio (local process) · Streamable HTTP
Runtime
Hosted (no local runtime) · Docker
Credentials
API key / credential required
License
Apache-2.0
Last commit
Tools
15
47FMRS · D

Beever Atlas is an innovative LLM knowledge base implementation that improves accuracy and traceability by distilling chat data into structured wikis and knowledge graphs. It offers extensive MCP integration and is suitable for teams needing knowledge management. However, it requires self-hosting, configuration is complex, and API stability is not guaranteed, making it best for technically capable teams.

Strongest · Documentation 14/20 Weakest · Reliability 5/20

Reliability
5/20
Security and permissions
10/20
Maintenance
10/20
Documentation
14/20
Setup experience
8/20
Why each score
Reliability 5/20
The manifest is well-formed and its entry point (python -m beever_atlas.api.mcp_server) matches the stdio module named in the README, which is a small positive; however, the README explicitly states this stdio mode has no HTTP server and no backing stores, meaning the packaged OCI/stdio entry can only serve tool-list introspection, not real knowledge-base queries. The primary documented client path is streamable-http, contradicting the manifest's stdio transport; the claimed 28-tool catalog, the init/tool-list handshake, and whether declared tools match real behavior are unverifiable from the supplied source, with no CI or committed tests provided, so the static-review cap keeps this score low.
Security and permissions 10/20
No red-line triggers: no malware, no credential theft or covert exfiltration patterns, and install examples use placeholders rather than real tokens. The README discloses MCP auth via BEEVER_MCP_API_KEYS, principal-keyed rate limits, AES-256-GCM encrypted storage of platform credentials under CREDENTIAL_MASTER_KEY, a no-telemetry claim, and fail-fast startup against dev defaults in production — all positive signals; however, the tool catalog includes long-running operations whose side effects (e.g., triggering syncs or paid LLM calls) and least-privilege scoping cannot be confirmed from source, so the score rests at the 10 anchor rather than higher.
Maintenance 10/20
Apache-2.0 license, SECURITY.md, CODE_OF_CONDUCT, NOTICE, a documentation site, and a versioned release (0.3.0) are all present and the repository is not archived, giving a solid ownership baseline; but this review can only inspect the manifest and README, so commit cadence, issue responsiveness, and dependency-update practices are unverifiable. With 51 open issues and an API-stability section still describing a 0.1.0/0.2.0 roadmap while the manifest is 0.3.0, there is visible governance/versioning lag, which matches the 10-point anchor exactly.
Documentation 14/20
The README is layered and thorough: three install options, a key-purpose table, a production-hardening checklist with reasons, architecture explanation, demo mode, common commands, and a dedicated docs/mcp-server.md covering the tool catalog, auth setup, .mcp.json templates for Claude Code/Cursor, and rate limits; privacy and API-stability caveats are also disclosed. Deductions apply for the self-contradictory description of the packaged stdio mode (it claims to expose the same tool catalog yet has no backing stores, misleading registry users into expecting real queries), stale API-stability text citing 0.1.0/0.2.0 against a 0.3.0 manifest, and thin cost/troubleshooting detail.
Setup experience 8/20
The full deployment path is well documented: docker compose with seven services, a guided ./atlas installer with non-interactive mode, secret-generation commands, and client .mcp.json examples for Claude Code. However, the reviewed server.json packages a docker/stdio entry that the README describes as having no backing stores — suitable only for registry introspection, not a working connection to a real knowledge base. Functional MCP use requires standing up the entire stack first, including two mandatory API keys and generated secrets, which is heavy and manual; with no CI/test evidence in the supplied material, the score sits conservatively below the static cap of 15.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessUses the networkConnects to a database

Best for

  • Teams that need to distill large volumes of chat into searchable, citable knowledge.
  • RAG applications that want to improve answer consistency and reduce hallucinations.
  • Developers who want MCP integration to let AI coding assistants access team knowledge.

Not for

  • Lightweight use cases that just need a simple chat log search.
  • Users who cannot self-host Docker-based infrastructure or prefer serverless options.
  • Those needing deep integration with proprietary knowledge management products without self-hosting.

Required permissions

  • Requires read access to Slack, Discord, Teams, or Mattermost channels (bot tokens configured via UI).
  • Access to external APIs: Gemini, Jina Embeddings (required); optional Tavily/Olostep web search.
  • Runtime requires Docker, network ports (3000, 8000, 3001, 8080, 7474, 7687, 27017, 6380).
  • For MCP clients, requires BEEVER_MCP_API_KEYS authentication.

Risks and side effects

  • Indexed chat data may contain sensitive information; ensure data security in deployment.
  • API keys (e.g., Gemini, Jina) if exposed can lead to misuse or data breaches.
  • Auto-generated wiki may contain errors or inaccurate info; human review may be needed.
  • Early-stage version (0.3.0) with unstable API; may break on updates.

Setup

Before you start

Runtime：Hosted (no local runtime) · Docker

BEEVER_MCP_KEY requiredsecret Bearer token for the MCP server, generated via BEEVER_MCP_API_KEYS management, see docs/mcp-server.md.
GOOGLE_API_KEY requiredsecret Gemini API key for extraction, entity graph, and answers; get it at aistudio.google.com/apikey.
JINA_API_KEY requiredsecret Jina v4 embeddings (2048-dim) key for semantic search; get it at jina.ai/api-dashboard.
TAVILY_API_KEY optionalsecret Optional external web search when QA retrieval confidence is low; get it at tavily.com.
OLOSTEP_API_KEY optionalsecret Optional web search alternative to Tavily; set WEB_SEARCH_PROVIDER=olostep, get it at olostep.com/dashboard.
  1. Clone the repo and enter: git clone https://github.com/beever-ai/beever-atlas.git && cd beever-atlas.
  2. Optional: run make demo to try the seeded demo (no API keys needed).
  3. Obtain API keys: GOOGLE_API_KEY (Gemini) and JINA_API_KEY (embeddings) are required; optionally TAVILY_API_KEY or OLOSTEP_API_KEY for web search.
  4. Choose deployment: recommended ./atlas interactive installer, or manually copy .env.example and fill in keys, then run docker compose up -d --build.
  5. Open the dashboard (default http://localhost:3000) and configure AI providers in Settings.
  6. Add a workspace (Slack/Discord/Teams) under Connections and sync channels.
  7. For MCP clients, configure .mcp.json with streamable-http or stdio as per docs.
.cursor/mcp.json 
{
  "mcpServers": {
    "beever-atlas": {
      "url": "https://atlas.example.com/mcp",
      "transport": "streamable-http",
      "headers": {
        "Authorization": "Bearer ${BEEVER_MCP_KEY}"
      }
    }
  }
}

Shown for Cursor. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "beever-atlas": {
      "url": "https://atlas.example.com/mcp",
      "transport": "streamable-http",
      "headers": {
        "Authorization": "Bearer ${BEEVER_MCP_KEY}"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add --transport http beever-atlas https://atlas.example.com/mcp --header 'Authorization: Bearer ${BEEVER_MCP_KEY}'

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After connecting the client at https://atlas.example.com/mcp with the Bearer ${BEEVER_MCP_KEY} header, run list_tools — the 15+ recorded tools (search_facts, get_channel_wiki, start_sync, etc.) should appear — then call list_channels to confirm the auth key works against your workspace.

Troubleshooting

  1. If containers fail, check `docker compose logs -f beever-atlas` for backend logs.
  2. If sync doesn't work, verify platform credentials are correctly entered in the UI.
  3. If MCP connection fails, check BEEVER_MCP_API_KEYS, URL, and auth headers.
  4. If QA quality is poor, confirm GOOGLE_API_KEY and JINA_API_KEY are valid and models available.
  5. If port conflicts, adjust port settings in .env.

Things to try

Once connected, you can ask your AI assistant things like:

  • List the channels available in my Beever Atlas workspace and show the wiki for the most recently synced one.
  • Search the knowledge base for facts about our deployment decisions and cite the source messages.
  • Use get_relations to show which people are connected to the payments project in the entity graph.
  • Ask the knowledge base: what did we decide about the API rate limits last month?

Tools 15

list_tools read-only
List all available tools on the server.
search_facts read-only
Search atomic facts in the knowledge base.
search_topics read-only
Search topic pages.
get_topic read-only
Get details for a specific topic.
get_channel_wiki read-only
Get the auto-generated wiki for a channel.
get_channel_summary read-only
Get a summary or overview of a channel.
get_graph read-only
Get entity-relationship view of the knowledge graph.
get_relations read-only
Query relations for a specific entity.
Show 7 more tools
get_qa read-only
Ask the Q&A agent a question and get a cited answer.
start_sync writes
Start a channel sync.
get_sync_status read-only
Get the status of a sync operation.
list_channels read-only
List available channels.
list_workspaces read-only
List connected workspaces.
search_entities read-only
Search entities (people, projects, etc.).
get_entity read-only
Get details for a specific entity.

Use cases

Automatically turn team chat history into a structured wiki for onboarding and knowledge sharing.
Let AI agents like Claude Code and Cursor query team knowledge via MCP and get cited answers.
Answer relational questions across channels, such as 'who worked on X with Y?'.
Track decisions, project progress, and personnel in a searchable knowledge base.

Supported clients

Claude
Cursor

Listed from the project's documentation, not tested by this site.

Overview

Beever Atlas is an open-source LLM knowledge base that connects Slack, Discord, Microsoft Teams, and Mattermost. It extracts atomic facts from messages, deduplicates them, and clusters them into topic pages with citations. Data is stored in a semantic store (Weaviate) and a graph store (Neo4j), enabling relational queries about people, decisions, and projects. Users can query the knowledge base via a dashboard or through MCP, allowing AI agents like Claude Code and Cursor to get cited answers in natural language.

Similar servers

Telegram MCP Server 70 · B

Telegram MCP server powered by Telethon to let MCP clients read chats, manage groups, and send/modify messages, media, contacts, and settings.

★ 1.8k · Tools 76 Compare with this →
Compare with other MCP servers

Source revision 7d791af27ef4 Data synced 2026-10-12 Read the FMRS scoring method