
Open Brain
A graph-based memory layer that scores 37.2% on LongMemEval and actually publishes the benchmark. Captures thoughts from Telegram, RSS, or direct API calls, then runs automated deduplication, theme tracking, and decay cycles in the background. Built on Postgres with pgvector for semantic search. Exposes 17 MCP tools for searching by meaning, browsing by metadata filters, and pulling weekly theme reviews. Named entities get resolved into a shared graph with typed relationships like extends, contradicts, or supersedes. Co-occurrence edges strengthen based on retrieval patterns. Self-host with Docker Compose or deploy to Supabase with included scripts. Good fit if you want persistent memory across conversations without manual tagging or folder hierarchies.
- tools
- 16
- auth
- Required
- transport
- HTTP
- last updated
- Apr 16, 2026
Tools
16 toolsPublic tool metadata for what this MCP can expose to an agent.
search_thoughtsSemantic search with optional graph expansion (1-hop traversal)
Semantic search with optional graph expansion (1-hop traversal)
No parameter schema in public metadata yet.
list_thoughtsBrowse thoughts filtered by type, topic, person, theme, quality, time
Browse thoughts filtered by type, topic, person, theme, quality, time
No parameter schema in public metadata yet.
thought_statsAggregate statistics: counts, type/theme breakdown, top topics/people
Aggregate statistics: counts, type/theme breakdown, top topics/people
No parameter schema in public metadata yet.
capture_thoughtSave a new thought from any AI client (with auto-embedding)
Save a new thought from any AI client (with auto-embedding)
No parameter schema in public metadata yet.
get_connectionsGraph traversal from a thought (typed links: extends, contradicts, etc.)
Graph traversal from a thought (typed links: extends, contradicts, etc.)
No parameter schema in public metadata yet.
list_entitiesBrowse extracted entities (people, tools, projects, orgs) by frequency
Browse extracted entities (people, tools, projects, orgs) by frequency
No parameter schema in public metadata yet.
weekly_reviewAI-generated summary of themes, open loops, and next steps
AI-generated summary of themes, open loops, and next steps
No parameter schema in public metadata yet.
analyzeGraph analysis: hubs, density, sources, co-occurrence, themes
Graph analysis: hubs, density, sources, co-occurrence, themes
No parameter schema in public metadata yet.
dedup_reviewDuplicate candidates with similarity zone histogram
Duplicate candidates with similarity zone histogram
No parameter schema in public metadata yet.
refresh_salienceRecompute all salience scores
Recompute all salience scores
No parameter schema in public metadata yet.
update_thoughtRewrite content (re-embeds, re-extracts metadata)
Rewrite content (re-embeds, re-extracts metadata)
No parameter schema in public metadata yet.
delete_thoughtPermanent delete (cascades connections)
Permanent delete (cascades connections)
No parameter schema in public metadata yet.
serendipity_digestResurface forgotten high-quality thoughts
Resurface forgotten high-quality thoughts
No parameter schema in public metadata yet.
pipelinePipeline monitoring: health status, run history, merge audit
Pipeline monitoring: health status, run history, merge audit
No parameter schema in public metadata yet.
review_staleReview and act on stale thought candidates
Review and act on stale thought candidates
No parameter schema in public metadata yet.
migration_guideInstructions for importing memories from other platforms
Instructions for importing memories from other platforms
No parameter schema in public metadata yet.
Sponsors
Readme
Open Brain
Graph-structured MCP memory. 37.2% on LongMemEval baseline — a benchmark most memory systems don't publish.
A self-hostable memory server for MCP clients (Claude, ChatGPT, any assistant that speaks MCP). Thoughts flow in from Telegram, pipelines, or direct capture, and land in a Newman-IDF weighted entity graph — not a flat document store. An automated Dream cycle runs in the background: deduplicating near-duplicates, tracking theme drift, synthesizing insights across clusters, and archiving stale content. 17 MCP tools. PostgreSQL + pgvector. You own your data.
How It Works
flowchart LR
TG[Telegram Message] --> TGBot[telegram-bot\nEdge Function]
MCP[AI Client\nClaude/ChatGPT] --> MCPServer[open-brain-mcp\nEdge Function]
Pipeline[RSS/HF Papers/\nEmergent Mind] --> RunPipeline[run-pipeline\nEdge Function]
TGBot --> OR1[OpenRouter\nEmbedding + Metadata]
MCPServer --> OR2[OpenRouter\nEmbedding + Search]
RunPipeline --> OR3[OpenRouter\nTriage + Embed]
OR1 --> DB[(Postgres\n+ pgvector)]
OR2 --> DB
OR3 --> DB
MCPServer --> DB
TGBot --> TGReply[Telegram Reply\nwith Metadata]
Details
- Repository
- github.com/bobby-cell-commits/open-brain-server
- Category
- AI & LLM Tools
- Transport
- HTTP
- Auth
- Required
- Registry
- MCP registry, active