
fish-bridge — Session Knowledge Graph
Turns sprawling AI chat sessions into compact knowledge graphs that persist across conversations. The core workflow runs standalone: ingest from Copilot JSONL or paste any chat, compile to a 300-800 token graph, write to copilot-instructions.md or CLAUDE.md. The optional MCP server adds real-time turn recording in VS Code agent mode via record_turn, get_context, mark_resolved, and export_session tools. Supports local Ollama (free, offline) or cloud backends like Gemini and Claude. Useful when you're resuming long-running architecture discussions or onboarding someone mid-project. Also merges external sources like Obsidian vaults, git logs, dependency manifests, and test output into the same typed graph structure.
- GitHub stars
- 1
- tools
- 9
- auth
- Secret needed
- transport
- STDIO
- env vars
- 3
- last updated
- May 17, 2026
Tools
9 toolsPublic tool metadata for what this MCP can expose to an agent.
record_turn
No parameter schema in public metadata yet.
get_context
No parameter schema in public metadata yet.
get_focus
No parameter schema in public metadata yet.
mark_resolved
No parameter schema in public metadata yet.
add_node
No parameter schema in public metadata yet.
export_session
No parameter schema in public metadata yet.
import_session
No parameter schema in public metadata yet.
show_active
No parameter schema in public metadata yet.
list_deferred
No parameter schema in public metadata yet.
Sponsors
Configuration
ANTHROPIC_API_KEYsecretAPI key for Claude extraction backend (omit when using local/Ollama)
OPENAI_API_KEYsecretAPI key for OpenAI extraction backend (omit when using local/Ollama)
GEMINI_API_KEYsecretAPI key for Gemini extraction backend (omit when using local/Ollama)
Readme
fish_bridge
Session-scoped knowledge graph engine for AI chat context compression.
Converts raw AI chat (40k+ tokens) into a compact typed knowledge graph (~300–800 tokens) and writes it to .github/copilot-instructions.md or CLAUDE.md — automatically included in every AI turn across all modes (ask, edit, agent). No MCP server required for the core workflow.
Raw session (40k tokens) → [fish_bridge] → Compressed graph (350 tokens)
written to copilot-instructions.md
picked up by every AI turn automatically
Install
Details
- Repository
- github.com/makeamouse/fish-bridge-mcp
- Categories
- AI & LLM Tools, Communication & Messaging, Documents & Knowledge
- Package
fish-bridge-mcp- Transport
- STDIO
- Auth
- Required
- Registry
- MCP registry, active