Fish Bridge — Session Knowledge Graph
Turn long Claude or Cursor chat threads into a compact, typed knowledge graph your agent can query instead of re-reading thousands of tokens of history.
Overview
fish-bridge MCP is a MCP server for the build phase that compresses AI chat sessions into a typed knowledge graph with optional Ollama-backed local extraction.
What is this MCP server?
- Compresses AI chat sessions into a typed knowledge graph for later agent retrieval
- Supports local/offline extraction via Ollama without cloud LLM keys
- Optional cloud extraction backends: Anthropic, OpenAI, or Gemini API keys in server env
- PyPI stdio package fish-bridge-mcp at manifest version 0.1.3
- Session memory MCP—not a generic notes app or vector DB host
- 1 PyPI stdio package (identifier fish-bridge-mcp)
- Up to 3 optional secret LLM API environment variables (Anthropic, OpenAI, Gemini)
Community signal: 1 GitHub stars.
What problem does it solve?
Your agent sessions grow until you lose decisions in scrollback, burn tokens re-explaining context, and have no structured memory to carry from yesterday’s debug thread into today’s feature work.
Who is it for?
Power users running long multi-day agent threads who want offline-friendly compression with Ollama or optional Claude/OpenAI/Gemini extraction.
Skip if: Builders who only need occasional chat export to markdown or who want a hosted team knowledge base without running Python MCP and an extraction backend.
What do I get? / Deliverables
Past chats become a queryable knowledge graph inside MCP so follow-up tasks start from structured session memory instead of raw transcript dumps.
- Typed knowledge graph derived from prior agent chat sessions
- Configured extraction backend (local Ollama or cloud API)
- Reusable structured context for follow-up agent tasks without full transcript replay
Recommended MCP Servers
Journey fit
Agent-heavy builds generate enormous session context; compressing it into structured memory is core tooling for anyone shipping with LLM assistants daily. Agent-tooling is where session bridges, memory layers, and offline extraction backends belong in the Skillselion journey—not a one-off ship checklist.
How it compares
Session-to-knowledge-graph MCP bridge—not RAG over your whole monorepo and not a code review or planning skill.
Common Questions / FAQ
Who is fish-bridge MCP for?
Indie developers and agent-first builders who accumulate long Claude Code or Cursor sessions and need structured, reusable memory without paying to re-send entire transcripts each time.
When should I use fish-bridge MCP?
Use it after dense debugging or planning threads, when switching tasks in the same product, or when you want local Ollama extraction instead of shipping chat logs to a cloud model.
How do I add fish-bridge MCP to my agent?
Add the PyPI stdio server fish-bridge-mcp to your MCP configuration, install Python dependencies per the repo, set optional ANTHROPIC_API_KEY, OPENAI_API_KEY, or GEMINI_API_KEY—or configure Ollama for local mode—and restart your agent client.