# leakferret [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/leakferrethq/leakferret  
**GitHub Stars:** 4  
**npm Downloads (last month):** 291  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/leakferret

## Description
Context-aware secret scanner: lets an AI agent scan, verify, and rewrite secrets before committing.

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

```json
"mcpServers": {
  "leakferret": {
    "command": "npx",
    "args": ["-y","@leakferret/mcp"]
  }
}
```

## Documentation

## What leakferret MCP server does

The leakferret MCP server exposes leakferret’s local secret-scanning engine to coding agents. It checks repository files for more than 60 secret patterns, including dotfiles such as `.env`, while respecting `.gitignore`. A separate option can scan Git history rather than only the working tree.

The server supports a workflow that moves from detection to decision and remediation. Candidates can be classified as real, fixture, or unknown; known public examples and dummy values can therefore be excluded from actionable results. Provider verification can then determine whether a credential is live. Rewrite proposals replace hardcoded literals with language-appropriate environment-variable lookups and can include an `.env.example` entry and secret-manager seed commands.

## How it works

The MCP process runs locally and communicates with the agent using JSON-RPC over stdio. The raw secret is kept on the local machine. Reports, logs, and network messages contain only redacted previews, while provider verification calls go directly from the machine to the relevant provider.

Five MCP tools cover the main workflow:

- `scan_repository` finds candidate secrets in a repository.
- `classify_candidates` assigns classifications using offline rules and heuristics, with model-assisted classification available through the supplied prompt.
- `verify_finding` makes a provider check to determine whether a finding is live.
- `propose_rewrite` prepares an environment-variable rewrite for a hardcoded value.
- `baseline_diff` compares findings with a baseline containing one-way HMAC fingerprints, allowing CI to focus on new findings.

The engine natively verifies credentials for roughly 25 providers, including AWS, GitHub, GitLab, Stripe, OpenAI, Anthropic, Slack, Twilio, SendGrid, and others listed by the project. A trufflehog binary can provide fallback verification for additional credential types.

## Setup and configuration

Run the packaged MCP launcher with `npx @leakferret/mcp`, then register it under `mcpServers`. A native leakferret binary can instead be started with the `mcp` argument. Claude Desktop, Cursor, Continue, and Claude Code are shown as supported configuration targets in the project documentation. Claude Code can use a project-level `.mcp.json` file or its `claude mcp add` command.

The optional `LEAKFERRET_BIN` environment variable points wrappers at a local binary, which is useful for offline or development installations. No server-side account or separate verification service is described. The provider APIs receive verification requests directly from the local machine.

## Tools and capabilities

In addition to tools, the leakferret MCP server provides a `classify` prompt for inline model-assisted classification. Two read-only resources expose the bundled engine metadata: `leakferret://secret-types` lists detectable patterns, and `leakferret://verifiers` lists live-verification providers.

The server is a fit for agent-driven development workflows where generated code may introduce credentials. It can help an agent inspect a repository, verify whether a candidate is active, suggest a safe environment lookup, and identify newly introduced findings through a baseline.

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