Definitive annual report on secret sprawl. 81% surge in AI-service key leaks. 24,008 secrets found in MCP configurations. Quantifies the problem that secret scanning MCP tools solve.
GitGuardian MCP (ggmcp)
activeBest purpose-built secret scanning MCP. 500+ detectors. Hard merge gate for AI-generated code. State of Secrets Sprawl 2026 report documents 81% surge in AI-service key leaks and 24,008 secrets found in MCP configs.

Where it wins
500+ secret detectors — broadest coverage in the category
Hard merge gate for AI-generated code — blocks secrets before they land
State of Secrets Sprawl 2026 report — definitive industry research on secret leaks
81% surge in AI-service key leaks documented — proves the problem this tool solves
24,008 secrets found in MCP configs — directly relevant to agent security
Official MCP server (ggmcp) — first-class agent integration
Where to be skeptical
Commercial platform — free tier exists but full features require subscription
MCP server community size smaller than Gitleaks or TruffleHog OSS projects
Editorial verdict
#1 secret detection. Purpose-built secret scanning MCP with 500+ detectors and hard merge gates for AI-generated code. State of Secrets Sprawl 2026 report (81% surge in AI-service key leaks, 24,008 secrets in MCP configs) is the definitive source on the problem. The default recommendation for secret scanning in agent workflows.
Related

CodeQL (via GitHub MCP Server)
88GitHub-native SAST via CodeQL, accessible through the official GitHub MCP Server. Copilot Autofix generates fixes from CodeQL alerts. GitHub Security Lab Taskflow Agent found ~30 real CVEs. Zero extra setup for GitHub users.

TruffleHog
8818K+ stars. 800+ secret types. Unique credential verification — confirms if leaked creds are still active. Scans S3, Docker, Slack — not just git. No official MCP but community integrations exist.
Gitleaks
8824.4K stars — most-starred secret scanner. 150+ patterns. Fastest pre-commit scanner. The community default for pre-commit secret detection. No official MCP.

Tencent AI-Infra-Guard
81Most comprehensive OSS AI red teaming tool. 3,264 stars. Full-stack: ClawScan, Agent Scan, Skills Scan, MCP scan, jailbreak eval. 43 AI framework components, 589 CVEs cataloged. v4.0 released.
Public evidence
Official MCP server for GitGuardian. 500+ secret detectors. Hard merge gate blocks secrets in AI-generated code before they reach the repository.
Raw GitHub source
GitHub README peek
Constrained peek so you can sanity-check the source material without leaving the site.
GitGuardian MCP Server
Bring GitGuardian's secret detection and incident management into your AI agent. Scan code for credentials before they leak, triage existing incidents, generate honeytokens, and remediate findings — all from inside your IDE or chat client, backed by GitGuardian's 500+ detectors.
[!CAUTION] MCP servers are an emerging technology. Agents act on your behalf and under your responsibility. Use trusted MCP servers and review agent actions when they interact with tools. To limit blast radius the server defaults to read-only-leaning permissions; what is actually exposed is determined by the OAuth scopes your access token holds.
What it does
- Secret scanning — proactively scan files for leaked credentials.
- Incident management — list, filter, assign, resolve, and tag incidents (both internal and Public Monitoring incidents).
- Honeytokens — generate honeytokens and list existing ones.
- Code-fix automation — open pull requests that remediate secrets in repositories your workspace monitors.
The exact set of tools exposed to your agent depends on the OAuth scopes granted to your access token.
Prompt examples
Scan this codebase for any leaked secrets or credentials.
Remediate all incidents related to my project.
Check if there are any new security incidents assigned to me.
Help me understand this security incident and provide remediation steps.
List all my active honeytokens.
Generate a new honeytoken for monitoring AWS credential access.
Create a honeytoken named 'dev-database' and hide it in config files.
Quick start
The recommended way to run the GitGuardian MCP server is to point your MCP
client at the hosted server. The MCP client handles OAuth automatically; no
local install, no token to manage, no uvx.
Pick the URL that matches your GitGuardian region:
| Region | URL |
|---|---|
| US SaaS | https://mcp.gitguardian.com/mcp |
| EU SaaS | https://mcp.eu1.gitguardian.com/mcp |
| Self-hosted | See Self-hosting the MCP server |
Cursor
Edit ~/.cursor/mcp.json:
{
"mcpServers": {
"GitGuardian": {
"type": "http",
"url": "https://mcp.gitguardian.com/mcp"
}
}
}
Claude Desktop
Edit ~/Library/Application Support/Claude Desktop/mcp.json (macOS) or
%APPDATA%\Claude Desktop\mcp.json (Windows). Same JSON as Cursor. Claude
Desktop versions that pre-date HTTP MCP support need the
Local stdio fallback.
Claude.ai (web)
Add the server in Settings → Connectors → Add custom connector with the URL above. OAuth is handled in the browser tab.
Windsurf
Edit ~/Library/Application Support/Windsurf/mcp.json (or
~/.config/Windsurf/mcp.json on Linux):
{
"mcp": {
"servers": {
"GitGuardian": {
"type": "http",
"url": "https://mcp.gitguardian.com/mcp"
}
}
}
}
Zed
Edit ~/Library/Application Support/Zed/mcp.json (or
~/.config/Zed/mcp.json on Linux) with the same type: http snippet.
Choosing a deployment
Two deployment paths are supported. Pick based on where your GitGuardian instance lives and what tradeoffs you accept.
| Deployment | When to use |
|---|---|
| Hosted MCP (Quick start above) | GitGuardian SaaS (US/EU) and you accept that requests transit mcp.gitguardian.com in addition to api.gitguardian.com |
| Self-hosted MCP (§) | Self-hosted GitGuardian, airgapped environments, or you want the MCP server on your own infrastructure |
| Local stdio with PAT (§) | CI/CD, scripts, one-off invocations, or older MCP clients without type: http support |
Authentication
Most users do not need to touch this — the Quick start config implicitly uses the OAuth proxy mode on the hosted server, and the Local stdio config uses PAT env.
There are four authentication modes the server can run in; you pick one via env vars.