Skip to content

Copilot / MCP Integration

The Model Context Protocol (MCP) lets AI agents like GitHub Copilot call external tools. IntentWeave exposes its index queries as MCP tools, so Copilot can search your code index, find connections, and check drift — all from chat.

Terminal window
iw mcp --session my-project -v

Add to .vscode/mcp.json:

{
"servers": {
"intentweave-kg": {
"command": "npx",
"args": ["@intentweave/cli", "mcp", "--session", "my-project", "-v"]
}
}
}

VS Code Copilot will auto-discover the server and make the tools available in chat.

These tools work with the local SQLite index — no Neo4j or LLM needed.

ToolPurposeKey Parameters
cari_retrieveRanked file retrieval by topic or symbolquery, scope?, limit?
cari_connectionsCross-layer connections + gap detectionentity, include?, limit?
cari_checkCI drift detection for changed fileschanged, severity?

Ask Copilot:

  • “Find files about authentication” → Copilot calls cari_retrieve with query=“authentication”
  • “What’s connected to AuthService?” → Copilot calls cari_connections with entity=“AuthService”
  • “I changed auth.ts — any docs to update?” → Copilot calls cari_check with changed files

If you have Neo4j running, additional tools are available:

ToolPurposeKey Parameters
kg_queryNatural language or Cypher queryquestion, cypher?, limit?
kg_contextBuild RAG context from graphtopic?, entity?, hops?
kg_entitiesList/search entitiestype?, search?, limit?
kg_impactSemantic impact analysisfiles, hops?
kg_doc_healthDocumentation freshnessfiles?
kg_schemaGraph schema description(none)
Terminal window
iw mcp [options]
OptionDefaultDescription
-s, --sessionDefault session scope
-v, --verboseoffLog tool invocations