Idea: Add init, scan, callgraph to tools

[christianwengert]

I think it would be great to be able to call drift init, drift scan and drift callback build to the MCP. Because my LLM has already access to our relevant repos so I do not need to call those manually from the command line. What do you think?

Thanks for your work!

[dadbodgeoff]

Love this idea @christianwengert and its just a natural evolution! I'll have this done today. Thanks for checking it out and giving this idea ill make sure to credit you :)

[christianwengert]

Thanks for having put that to work so quickly. I am currently testing it:

docker run --user=drift --env=PORT=3000 -p 3041:3000 --env=VERBOSE=true --env=PROJECT_ROOT=/workspace --volume=MYLOCALDATA:/workspace mcp-drift:latest

(BTW: I had to fix the Dockerfile, see #52).

Then the mcp server starts with

[drift-mcp-http] Starting server for: /workspace
[drift-mcp] Detected languages: none
[drift-mcp] Exposing 42/65 tools
[drift-mcp-http] Server running at http://0.0.0.0:3000
[drift-mcp-http] Project root: /workspace
[drift-mcp-http] Cache: enabled
[drift-mcp-http] Rate limiting: enabled
[drift-mcp] Warmup completed in 9ms
[drift-mcp] Loaded: {"patterns":0,"callGraph":false,"boundaries":false,"env":0,"dna":false,"contracts":0,"history":0,"coupling":false,"errorHandling":false,"testTopology":false}
[drift-mcp] Errors: Drift not initialized. Run `drift init` first.
[drift-mcp-http] New SSE connection: transport-1
[drift-mcp-http] New SSE connection: transport-2
...

This means my repos I want to check are in /workspace/USERID/repo_1 up to /workspace/USERID/repo_n.

I then use the mcp inspector (npx @modelcontextprotocol/inspector) to connect which works and it lists drift_setup as a tool (I had to change tool-filter.ts:123 and added the line 'drift_setup', otherwise the drift_setup tool would not show up).

drift_projects called with {}` returns

{
totalProjects:0
filteredCount:0
activeProject:null
projects:[]
}

drift_setup with action=status and no arguments gives

{
  "summary": "Drift not initialized in this project.",
  "data": {
    "success": true,
    "initialized": false,
    "projectPath": "/workspace",
    "detectedStack": {
      "languages": [],
      "orms": [],
      "frameworks": []
    },
    "hints": {
      "nextActions": [
        "Use drift_setup action=\"init\" to initialize drift",
        "Or use drift_setup action=\"full\" to init + scan + build callgraph"
      ]
    }
  },
  "meta": {
    "requestId": "req_ml6icejp_ta4e6h",
    "durationMs": 1,
    "cached": false,
    "tokenEstimate": 220
  },
  "hints": {
    "nextActions": [
      "Use drift_setup action=\"init\" to initialize drift",
      "Or use drift_setup action=\"full\" to init + scan + build callgraph"
    ]
  }
}

There is no active project, so "Drift not initialized in this project." makes probably sense. Now I would like to do the init to one of my repos with drift_setup with init:

{
  "action": "init",
  "project": "/workspace/USERID/repo_1"
}

returns:

{
  "error": {
    "code": "INTERNAL_ERROR",
    "message": "Internal error: Project \"/workspace/USERID/repo_1\" not found. Use drift_projects action=\"list\" to see available projects.",
    "recovery": {
      "suggestion": "This is an unexpected error. Please report it if it persists."
    }
  },
  "meta": {
    "requestId": "err_ml6im8cx",
    "timestamp": "2026-02-03T11:28:26.578Z"
  }
}

So I am currently a bit stuck. I tried to add the (non initialized project) using the drift_project command with these arguments:

{
  "action": "register",
  "project": "repo1",
  "path": "/workspace/USERID/repo_1"
}

Which results in

{
  "error": {
    "code": "INTERNAL_ERROR",
    "message": "Internal error: Project \"repo1\" not found. Use drift_projects action=\"list\" to see available projects.",
    "recovery": {
      "suggestion": "This is an unexpected error. Please report it if it persists."
    }
  },
  "meta": {
    "requestId": "err_ml6k6grz",
    "timestamp": "2026-02-03T12:12:10.223Z"
  }
}

I tested it with 956c6de

[dadbodgeoff]

hey! @christianwengert

Sorry I didnt get to fully look over and ensure it was fully 100% ready to go

Yesterday was a very hectic day... You'll see a commit that just went through that really revamps and makes cortex persistant memory id say one of the best in its class for what it does and its abilties truly now levels it up.

Also have been pushing to launch my cloud version and is a whole nother side prioject that hasnt been merged into the OSS yet...
As I think this could be very helpful for teams to set up how they want there agent to work and then have there teams set up the same parameter for each person coding with agents...Also with the addition of cortex a living bible of your memories that is recallable and viewable for easy parsing I think would be very cool

Let me look over all this stuff, test the PR and merge and get it fully finalized this morning.

Thanks for such a detailed report here and in the PR makes my life so much easier!

Much appreciated. Ill ping you back here in a bit.

[dadbodgeoff]

Hey @christianwengert Credits been given, added as a contributor through the repo and the fix should be in!
Let me know if theres any other issues or anything else I can do to make this better for you!

[christianwengert]

Thanks a lot! Great work. So far I was able to index a couple of projects. Now I will test a bit more the actual tools (signature, patterns, callers, etc).

Btw: Another question: Maybe we could also approve/disapprove patterns (or just autoapprove all) via MCP?

[dadbodgeoff]

Great idea @christianwengert just pushed this to main
Actions in drift_curate:

review - List patterns pending review with evidence requirements
verify - Verify a pattern exists with grep-based evidence (anti-hallucination)
approve - Approve a single pattern (requires evidence for non-high-confidence)
ignore - Ignore a pattern with a reason
bulk_approve - Auto-approve all high-confidence patterns (≥0.95)
audit - View curation decision history
Example usage via MCP:

// Review pending patterns
{ "action": "review", "minConfidence": 0.8 }

// Approve a specific pattern
{ "action": "approve", "patternId": "pattern-123" }

// Ignore a false positive
{ "action": "ignore", "patternId": "pattern-456", "ignoreReason": "Test code, not production" }

// Auto-approve all high-confidence patterns (dry run first)
{ "action": "bulk_approve", "dryRun": true }

// Actually bulk approve
{ "action": "bulk_approve", "confidenceThreshold": 0.95 }

Will update global tonight with all teh changes :)

Let me know how it works for you and thanks for the great idea!