Ogham Mcp
@ogham-mcp
Persistent shared memory for AI agents. Hybrid search (pgvector + tsvector), knowledge graph, cognitive scoring, and 16-language temporal extraction. 97.2% Recall@10 on LongMemEval with one PostgreSQL query. Works across Claude Code, Cursor, Codex, OpenClaw, and any MCP client
Overview
What is Ogham Mcp?
Ogham Mcp is a persistent, searchable shared memory server for AI coding agents. It enables agents across different client tools (Claude Code, Cursor, Kiro, OpenCode) to retain and search context across sessions. It achieves 97.2% recall@10 on the LongMemEval benchmark using a single PostgreSQL query.
How to use Ogham Mcp?
Install via pip (pip install ogham-mcp), Docker (ghcr.io/ogham-mcp/ogham-mcp), or integrate directly with Claude Code and OpenCode. Configure environment variables for database (Supabase, Neon, or vanilla Postgres), embedding providers, and temporal search. The server exposes MCP tools and can be run with SSE transport for multi-agent setups, or used through its CLI.
Key features of Ogham Mcp
- 97.2% retrieval recall@10 on LongMemEval benchmark
- Single PostgreSQL query – no LLM in search pipeline
- Persistent shared memory across different coding clients
- Tools for memory, search, graph, profiles, import/export
- Three built-in skills: ogham-research, ogham-recall, ogham-maintain
- Configurable embedding providers and temporal search
Use cases of Ogham Mcp
- AI coding agents sharing context across sessions and client tools
- Remembering past decisions, gotchas, and architectural patterns
- Reusing agent knowledge without re-explaining the codebase
- Multi-agent collaboration with shared memory via SSE transport
- Debugging recurring issues by retrieving relevant historical context
FAQ from Ogham Mcp
What runtime does Ogham Mcp require?
Ogham Mcp requires Python 3.13+ and a PostgreSQL database with pgvector and tsvector extensions.
How does Ogham Mcp achieve high retrieval accuracy?
It uses a hybrid search combining pgvector and tsvector (CCF) in a single SQL query, achieving 97.2% recall@10 without neural rerankers, knowledge graphs, or LLMs in the search loop.
What database options are supported?
Supabase, Neon, and vanilla PostgreSQL are supported. Full setup instructions are provided in the README for each option.
Is there a Docker image available?
Yes, a pre‑built Docker image is published at ghcr.io/ogham-mcp/ogham-mcp on GitHub Container Registry.
How can I install Ogham Mcp?
Install via pip (pip install ogham-mcp), Docker, or by integrating directly with MCP clients like Claude Code and OpenCode.
