Scan from your AI agent.
The MCPAmpel MCP plugin lets your agent discover its own installed servers and scan them in place. Install once. Ask Claude or Cursor "are my MCP servers safe?" — they call us, we tell them.
01 Install
Pick the package manager that matches how you ship Python. uvx is the recommended path — it runs the plugin in an isolated environment without polluting your system Python or your project's venv.
$ uvx mcpampel
$ pipx install mcpampel
$ pip install mcpampel
02 Configure your agent
Drop the server entry into your agent's MCP config. Restart the agent — the mcpampel tool surface appears in the next session.
{
"mcpServers": {
"mcpampel": {
"command": "uvx",
"args": ["mcpampel"]
}
}
}
{
"mcpServers": {
"mcpampel": {
"command": "uvx",
"args": ["mcpampel"],
"env": {
"MCPAMPEL_API_KEY": "sk_live_…"
}
}
}
}
{
"servers": {
"mcpampel": {
"transport": "stdio",
"command": "uvx",
"args": ["mcpampel"]
}
}
}
03 Use
The plugin exposes three tools. Your agent picks them automatically when the conversation calls for a security check.
Scan a specific URL
Submit any GitHub, GitLab, npm, or PyPI URL. Returns a trust score and per-engine findings.
Scan your installed servers
Reads the agent's MCP config, resolves each server to its source, scans them all in parallel.
Lookup an existing score
Returns the cached trust score for a known repo without spending a quota call. Sub-second.
04 Get an API key
Anonymous usage works for casual one-off scans. For consistent quota and your scan history, create a free account.
- 50 API calls / day on the free tier — enough for a CI hook plus daily agent use.
- Per-key analytics — see which servers your agent scanned, when, and the trust delta over time.
- No credit card. No upgrade nag.
Generate a key at /account and add it as MCPAMPEL_API_KEY in the env block of your MCP config.