# m1nd [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/maxkle1nz/m1nd  
**GitHub Stars:** 23  
**npm Downloads (last month):** 427  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/m1nd

## Description
The shell around your coding agent: a neuro-symbolic code graph with calibrated trust, via MCP.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "m1nd": {
    "command": "npx",
    "args": ["-y","@maxkle1nz/m1nd@1.6.5"]
  }
}
```

## Documentation

## What m1nd does

The m1nd MCP server gives coding agents a repository-specific structural model instead of requiring them to reconstruct relationships from repeated file searches. The graph represents where code lives, how symbols and files connect, and which findings have supporting evidence. It also retains knowledge between sessions, attaching conclusions to the code they describe and marking them when that code has changed.

Answers include a trust verdict. The system can report insufficient evidence, suggest verification, or identify a result that should not yet be trusted. Graph output remains guidance: the project explicitly requires file reads, tests, or runtime probes before a mission claim can be considered complete.

The local design keeps repository data on the developer’s machine. The project is distributed as one Rust binary and is licensed under MIT.

## How it works

Agents can orient themselves with `north`, investigate questions with `seek`, and inspect relationships through tools such as `impact`, `why`, and `ghost_edges`. Git history supplies “ghost edges” for files that repeatedly change together without an import relationship. Other analysis tools identify structural holes, recurring bug shapes, risky change velocity, and claims tested against the live graph.

Stored findings can be reinforced when later results confirm them, helping useful paths rank higher. Session presence lets multiple agents register work on the same repository. When overlapping work is detected, the agents receive a warning during orientation; the system does not decide which agent should proceed.

## Tools and capabilities

The documented MCP surface includes:

- `north` for repository orientation and trust information.
- `seek` for structural answers with a re-verification verdict.
- `impact`, `why`, `ghost_edges`, `missing`, `predict`, and `tremor` for dependency, coupling, absence, change-pattern, and risk analysis.
- `xray_gate` and `trust_selftest` for guardrail and trust checks.
- `antibody_scan` for searching the repository for previously identified structural bug patterns.
- `transplant_preview` and `transplant_commit` for guarded Rust symbol moves.

Transplant plans a move from a named symbol and destination, includes relevant documentation and attributes, classifies dependencies, updates referers, and returns a receipt. Commit rechecks planned file hashes and writes atomically. The operation reports unresolved references and state left behind rather than treating an incomplete result as success. A server-side boundary manifest can protect designated repository areas, with blocked operations failing without writing changes.

## Limitations and notes

The transplantation feature is limited to Rust, top-level functions, and moves within the same crate. The destination file must already exist, and references created inside macros are not visible to it.

Warnings and trust verdicts do not replace compilation, tests, or runtime validation. A graph-only result cannot close a mission. Multi-agent presence reports collisions but leaves the decision to the developer.

_Full upstream README: https://allmcps.com/mcp/m1nd/readme_