- Python 3.11+
- An LLM API key (Gemini, Anthropic, or local Ollama)
pip install fish-bridge-mcp
# or with a cloud backend extra (Gemini recommended)
pip install "fish-bridge-mcp[gemini]"
# See README for uv tool install (recommended) and other backends# Initialize for this project folder
fish-bridge init
# Configure your backend (set the API key as an environment variable)
export GEMINI_API_KEY=your-key-here
fish-bridge config --backend gemini
# or: --backend claude / --backend openai / --backend local
# Ingest the latest VS Code Copilot session (auto-discovered)
fish-bridge ingest --source copilot
# Or ingest from a saved export file:
fish-bridge ingest --source file --file path/to/export.json
# Compile context to your instructions file
fish-bridge compile --output .github/copilot-instructions.md| Command | What it ingests |
|---|---|
ingest --source copilot |
VS Code Copilot JSONL — auto-discovered from workspaceStorage |
ingest --source paste |
Any chat tool — opens $EDITOR, paste and save |
ingest --source file --file export.json |
From a saved export file (Claude JSON, plain text) |
merge --source document --file HANDOVER.md |
Markdown / text docs |
merge --source codebase --path . |
Git log + README |
merge --source obsidian --vault ~/notes |
Obsidian vault |
merge --source deps --path . |
package.json / pyproject.toml etc. |
merge --source testout --file results.json |
Jest/pytest/JUnit test output |
fish-bridge serve # opens http://localhost:8080 (Cytoscape.js)
fish-bridge show # pretty-print nodes in terminalfish-bridge serve-mcp # starts MCP server on stdioAdd to VS Code settings.json (see examples/copilot-mcp-config.json) or
Claude Desktop claude_desktop_config.json (see examples/claude-mcp-config.json).
- Configuration — all config file options
- Graph schema — node types, edge relations, status values
- Compiler modes — active thread, focus, digest
- Ingestors — format details for each ingest source