Agent quickstart

One address, nine tools, no account. Your agent can search, compare and install on its own.

Connect

The MCP server speaks Streamable HTTP and needs no sign-in.

MCPhttps://mcpvetted.com/api/mcp

Claude (web or desktop): Settings, Connectors, Add custom connector, paste the address.
Claude Code: claude mcp add --transport http mcpvetted https://mcpvetted.com/api/mcp
Cursor: one-click install, or add {"url": "https://mcpvetted.com/api/mcp"} to mcp.json.
VS Code: one-click install, or code --add-mcp.

The nine tools

Tool	Kind	What it does
`search_tools`	read	Searches the vetted index of MCP servers by keyword or by the job a tool does. Returns each result's trust tier, observed access, check results and trust page URL.
`get_tool`	read	Returns one indexed tool by slug, registry name or server URL: versions, tier history, observed access and supported apps.
`get_trust_report`	read	Returns each check's verdict and findings for one version of a tool, with the rules version and when each check ran.
`compare_tools`	read	Compares 2 to 5 tools side by side on tier, observed access, agent-reported outcomes and latest release.
`get_install_config`	read	Returns the exact install link, command or config block one app accepts for a tool, pinned to the vetted version, plus the access to show the user before installing. Records an install id for outcome reports.
`check_policy`	read	Checks one tool against a minimum tier and a list of blocked access types, and returns allow or deny with the reasons.
`list_revocations`	read	Lists tool versions revoked after failing a check, newest first, optionally since a date or for specific tools.
`submit_url`	write	Adds a tool to the index from an address: a remote MCP server URL, a domain, a GitHub repo, or an npm or PyPI package. Reads the server's self-description without calling its tools, then runs the standard checks.
`report_outcome`	write	Records whether a tool did its job after an install made with get_install_config. Reports from distinct agents feed the Proven tier.

Every tool declares an output schema and returns structured content plus a one-line summary. Full schemas: API reference.

A typical flow

search_tools with the job, e.g. schedule LinkedIn posts, and min_tier: "scanned".
compare_tools on the top two or three.
get_install_config for the user's app. Show the access list before installing, and keep the install_id.
After using the tool, report_outcome with that install_id. These reports build the Proven tier.

REST API

The same operations over HTTP, described in /openapi.json.

curl "https://mcpvetted.com/api/v1/search?q=calendar&min_tier=scanned&limit=3"
curl "https://mcpvetted.com/api/v1/tools/<slug>/trust"
curl "https://mcpvetted.com/api/v1/tools/<slug>/install?client=cursor"
curl -X POST "https://mcpvetted.com/api/v1/submit" -H "content-type: application/json" -d '{"url":"example.com/mcp"}'

Registry API for companies

Point VS Code and Copilot's MCP registry setting at https://mcpvetted.com/api/registry. It serves the MCP Registry v0.1 shape with only Scanned-and-up, unrevoked servers; our tier and check date ride in _meta["com.mcpvetted/trust"].

Rate limits and errors

Anonymous callers get 1,000 calls a day; an API key we issue (ask at the support address), sent in X-API-Key, raises that to 20,000. Submissions are limited to 20 an hour. Every response carries X-RateLimit-Limit, -Remaining and -Reset.

Errors say what was wrong and how to fix the call: REST returns { "error": { "code", "message", "hint" } }; MCP returns a tool result with isError: true and the same text.

For crawlers

/llms.txt summarizes the service, and every trust page has a Markdown twin at /t/<slug>.md.