Beever Atlas

What Beever-AI/beever-atlas MCP server does

The Beever-AI/beever-atlas MCP server turns conversations from Slack, Discord, Microsoft Teams, and Mattermost into a searchable team knowledge base. It extracts atomic facts, entities, relationships, topics, people, and decisions from synced messages. Those results populate an auto-generated wiki and a typed knowledge graph, with claims linked to the source messages used to produce them.

The MCP interface gives AI agents access to this stored knowledge through 28 tools. Agents can ask natural-language questions, perform semantic searches, retrieve cited answers, find subject-matter experts, and trace decisions. The same knowledge is also available through the Beever Atlas dashboard and browsable wiki.

This fits teams that need institutional knowledge from chat history to remain available after discussions move on. It is especially relevant for onboarding, locating who knows about a project, reviewing past decisions, and answering questions with citations back to conversations.

How it works

Supported chat histories pass through a multi-stage ingestion pipeline. The pipeline distils messages into atomic facts and extracts entities and relationships, while removing duplicate knowledge and grouping related information into topics.

Beever Atlas maintains two complementary memory layers. A three-tier semantic store organises channel, topic, and atomic-fact information for hybrid retrieval. A graph store connects people, projects, decisions, and other entities. The query system chooses between semantic and graph retrieval according to the question, then returns answers with citations to the originating messages.

Each synced channel receives an incrementally maintained wiki with sections such as topics, people, decisions, open questions, and timelines. The wiki is used as the primary retrieval surface, with raw messages available when a fact has not yet been distilled. The documented architecture uses Weaviate, Neo4j, MongoDB, and Redis behind backend, bot, and frontend services.

Setup and configuration

Beever-AI/beever-atlas is designed for self-hosted deployment and can run on-premises through Docker. The project supports bringing your own language model through LiteLLM rather than requiring a single model provider.

The server can be connected to Claude Code and Cursor as an MCP knowledge source.

Tools and capabilities

• Query the generated knowledge base in natural language.
• Search stored knowledge semantically.
• Return answers with citations to source messages.
• Find people associated with a topic or area of expertise.
• Trace decisions across conversations and linked entities.
• Expose the knowledge base through 28 MCP tools.

Limitations and notes

The Beever-AI/beever-atlas MCP server depends on prior ingestion: answers are limited by the conversations, imports, and channels that have been connected and synchronised. Newly discussed facts may not yet appear in the distilled wiki and can require a later sync or processing cycle.

The project is licensed under Apache 2.0. It belongs to the 🧠 Knowledge & Memory category because it builds searchable semantic and graph memory from team communications rather than acting as a general-purpose chat connector.