TrustedMCP MCP Server: Setup, Tools, Own Grade

Q: How do I add TrustedMCP to Claude Code?

Run claude mcp add --transport http trustedmcp https://trustedmcp.io/mcp --header "Authorization: Bearer YOUR_KEY" for the hosted server, or claude mcp add trustedmcp -e TRUSTEDMCP_API_KEY=YOUR_KEY -- npx -y trustedmcp mcp to run it locally.

On this page

TrustedMCP MCP Server: Grade Every Server in Your MCP Config (Nov 2026)

The TrustedMCP MCP server lets your agent check an MCP server: look up the A to D grade of any MCP server, check every server in your MCP config at once, find better-graded servers, and read official MCP Registry entries. Add it hosted at https://trustedmcp.io/mcp or run npx -y trustedmcp mcp, with a free key.

We built this server, so read this guide with that in mind. We grade it with the same rubric as every other listing, and its listing shows where it loses points. It never shows up as another server's alternative, and it isn't in our "best" lists.

What the TrustedMCP MCP server does

It has four read-only tools. Each one asks the TrustedMCP grades API and returns plain text your agent can quote.

Tool	Ask it	Key scope
`lookup_server`	"What grade does the Notion MCP server have?"	`servers:read`
`check_config`	"Check the MCP servers in my config."	`config:check`
`find_alternatives`	"Find a database MCP server graded B or better."	`servers:read`
`registry_lookup`	"Does the MCP Registry list any issues for this server?"	`registry:read`

lookup_server takes a name, an npm or PyPI package, a Docker image, a URL or a repo. check_config grades a whole config file. find_alternatives lists graded servers in a category, or servers graded better than the one you name. registry_lookup reads an entry in the official MCP Registry: its publisher, versions, how it runs, and any reported security issues.

None of the tools change anything. They read grades and registry data, and that's all.

Get a free API key

Every tool call needs a key. Get one at trustedmcp.io/developers with your email address.

Scopes. Pick the tools the key may use: servers:read, config:check, registry:read. A key without the right scope gets an error that names the missing one.
Revoke or replace it yourself. Paste the key on the developers page to revoke it or swap it for a new one.
Limits. 300 requests a minute per key. The hosted server and the JSON API share that limit. Without a key, your client can still connect and list the tools, but each tool call returns a message asking for a key.

Hosted or npx: which should you use?

Both run the same four tools and give the same answers.

	Hosted (`https://trustedmcp.io/mcp`)	Local (`npx -y trustedmcp mcp`)
Install	Nothing	Node.js 18 or later
Key	`Authorization: Bearer YOUR_KEY` header	`TRUSTEDMCP_API_KEY` env var
`check_config`	You paste the config into the chat	Reads your config files on your machine
Your config	Sent to trustedmcp.io over HTTPS	Stays on your machine

Use the npx version if you want check_config to find your config files on its own, or if you'd rather not paste a config into a chat. Use the hosted one for a quick try. Claude Desktop and claude.ai custom connectors can't send a header yet, so use the npx version there.

Add it to Claude Code

Hosted, no install:

claude mcp add --transport http trustedmcp https://trustedmcp.io/mcp \
  --header "Authorization: Bearer YOUR_KEY"

Or local:

claude mcp add trustedmcp -e TRUSTEDMCP_API_KEY=YOUR_KEY -- npx -y trustedmcp mcp

Run claude mcp list to see it. For the full set of Claude Code options, see Claude Code MCP setup.

Add it to Claude Desktop

Open claude_desktop_config.json (Settings, Developer, Edit Config). On macOS it's in ~/Library/Application Support/Claude/, on Windows in %APPDATA%\Claude\. Add:

{
  "mcpServers": {
    "trustedmcp": {
      "command": "npx",
      "args": ["-y", "trustedmcp", "mcp"],
      "env": { "TRUSTEDMCP_API_KEY": "YOUR_KEY" }
    }
  }
}

Quit and reopen Claude Desktop.

Add it to Cursor

In ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project), hosted:

{
  "mcpServers": {
    "trustedmcp": {
      "url": "https://trustedmcp.io/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

For the npx version, use the same block as Claude Desktop above.

Add it to VS Code

In .vscode/mcp.json, hosted:

{
  "servers": {
    "trustedmcp": {
      "type": "http",
      "url": "https://trustedmcp.io/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

Or local:

{
  "servers": {
    "trustedmcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "trustedmcp", "mcp"],
      "env": { "TRUSTEDMCP_API_KEY": "YOUR_KEY" }
    }
  }
}

If the file is committed to a shared repo, don't paste the key into it. VS Code can prompt for it instead: add an inputs entry with "type": "promptString" and "password": true, then use "${input:<id>}" as the value. More on VS Code's MCP support in VS Code MCP setup and security.

Add it to Windsurf

Open ~/.codeium/windsurf/mcp_config.json and add the same mcpServers block as Claude Desktop. Then refresh the MCP servers list in Windsurf's settings.

Check an MCP server from your agent: the four tools with real output

Each answer below is the tool's actual text, from the live API on 2026-11-12. Grades change when servers change, so yours may differ.

lookup_server: one server's grade

Prompt: "What grade does @playwright/mcp have?"

**Playwright MCP** (`playwright-mcp`): grade **C**
Category: Browser & Web Automation · Vendor: Microsoft
Drive a real browser to navigate, click, fill, and extract via MCP.
Checks:
- Login: no marks, no authentication
- Upkeep: full marks, actively maintained
- Known bugs: full marks, 0 open
- Connection: no marks, unsafe connection by default
Fix first: No login: keep it local and never expose it on a network.
Known CVEs: 0 · Upkeep: active · Last updated: 2026-10
Transport: stdio, http
Better-graded servers in the same category:
- Apify MCP Server (`apify-mcp-server`): grade A, https://trustedmcp.io/servers/apify-mcp-server
Page: https://trustedmcp.io/servers/playwright-mcp

A C here doesn't mean "don't use it". Playwright MCP is well maintained. It loses its marks because it has no login and can listen over HTTP. Run it over stdio on your own machine and that risk mostly goes away. The "Fix first" line tells you what to change.

check_config: every server in your config

Prompt: "Check the MCP servers in my config." With the npx version and no arguments, the tool finds the Claude Desktop, Claude Code, Cursor, VS Code and Windsurf configs on your machine. Here we gave it this config instead:

{
  "mcpServers": {
    "github": { "command": "docker", "args": ["run", "-i", "--rm", "-e", "GITHUB_PERSONAL_ACCESS_TOKEN", "ghcr.io/github/github-mcp-server"],
                "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "YOUR_TOKEN" } },
    "playwright": { "command": "npx", "args": ["@playwright/mcp@latest"] },
    "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/projects"] },
    "notion": { "url": "https://mcp.notion.com/mcp" },
    "my-notes": { "command": "node", "args": ["/Users/me/tools/notes-mcp/index.js"] }
  }
}

Checked 5 MCP servers in 1 config: 4 graded (A 1, B 1, C 1, D 1), 0 only in the official registry, 1 not found.

### pasted config
- **github**: grade **A**, GitHub MCP Server. Fix first: Login is coarse: sign in with the narrowest account or scopes you can. https://trustedmcp.io/servers/github-mcp-server
- **playwright**: grade **C**, Playwright MCP. Fix first: No login: keep it local and never expose it on a network. Better graded in its category: Apify MCP Server (A). https://trustedmcp.io/servers/playwright-mcp
- **filesystem**: grade **D**, Filesystem MCP Server. Fix first: No login: keep it local and never expose it on a network. Better graded in its category: Appwrite MCP Server (A), Buildkite MCP Server (A), Sentry MCP Server (A). https://trustedmcp.io/servers/filesystem-mcp-server
- **notion**: grade **B**, Notion MCP. Fix first: Login is coarse: sign in with the narrowest account or scopes you can. https://trustedmcp.io/servers/notion-mcp
- **my-notes**: runs `node` with a local file or a name we can't match, so it can't be looked up. Review it yourself.

To ask for a grade: https://trustedmcp.io/submit

The token value YOUR_TOKEN was never sent. Only the image name, the package names and the Notion URL were. The "better graded" names are servers in the same category with a higher grade. They don't always do the same job, so check that one fits before you swap.

find_alternatives: better-graded servers

Prompt: "Find something better graded than browser-use."

browser-use MCP is graded C, in Browser & Web Automation. 1 graded server at B or better in Browser & Web Automation; top 1:
1. **Apify MCP Server** (`apify-mcp-server`): grade **A**, Browser & Web Automation. Fix first: Login is coarse: sign in with the narrowest account or scopes you can. https://trustedmcp.io/servers/apify-mcp-server

You can also pass a category, a search term, a minimum grade (B by default) and a limit of up to 20.

registry_lookup: an official MCP Registry entry

Prompt: "Does the MCP Registry list any issues for io.github.github/github-mcp-server?"

**GitHub** (`io.github.github/github-mcp-server`) in the official MCP Registry
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
Listed version: 2.1.0 · Runs as: remote, oci · Status: active · Versions tracked: 63
Package: oci ghcr.io/github/github-mcp-server:2.1.0
Remote: https://api.githubcopilot.com/mcp/ (streamable-http)
Repo: https://github.com/github/github-mcp-server
TrustedMCP grade: A (`github-mcp-server`), https://trustedmcp.io/servers/github-mcp-server
Reported issues (2 reported issues, listed version not flagged):
- CVE-2026-47427: ... Prior to 1.1.0, the CompletionsHandler function ... [high, listed version not checked, published 2026-07-28]
- CVE-2026-48529: ... From 0.22.0 until 1.1.2, when running in HTTP mode with --lockdown-mode enabled ... [medium, listed version not checked, published 2026-06-26]
Page: https://trustedmcp.io/registry/io.github.github/github-mcp-server

We shortened the two CVE lines. Both were fixed in 1.1.x, well before the listed 2.1.0. Most registry entries have no TrustedMCP grade, and this tool still shows their packages, remotes and reported issues.

What leaves your machine

With the npx version, check_config reads your config on your machine and sends only:

server names from the config,
npm and PyPI package names,
Docker image names,
server URLs, without query strings.

Env values, headers, tokens, other arguments, file paths and the file itself stay on your machine. The source is on GitHub under the MIT license, so you can check that in src/config.ts.

With the hosted server, you paste the config into the chat, so the whole text goes to trustedmcp.io over HTTPS. We read only the same fields and ignore the rest. We don't store or log what you send. If that's not good enough for a config with real tokens in it, use the npx version or remove the env blocks before you paste.

Your key is tied to the email you signed up with. Each request counts toward that key's limit.

Its own grade, and where it loses points

TrustedMCP grades its own server A (see the listing and how we score servers). Like every grade here, it's an assessment from public signals, not an audit.

Login (full marks): from version 0.2.0, every tool call needs a key with scopes, and you can revoke it. Only the handshake and the tool list answer without one.
Upkeep (lost points): it's new. The first release was 2026-11-11, and one person maintains it. That's the biggest deduction, and the reason the "Fix first" line says to pin the version you tested, for example npx -y trustedmcp@0.2.0 mcp.
Known bugs (full marks): no advisories in the repo, and nothing in OSV for the npm package.
Connection (full marks): HTTPS for the hosted server, stdio for the npx one, read-only tools, and no shell commands.

You can check the same answer from your agent (shortened):

**TrustedMCP MCP Server** (`trustedmcp-mcp-server`): grade **A**
Checks:
- Login: full marks, scoped login by default
- Upkeep: part marks, updated recently, upkeep hard to verify
- Known bugs: full marks, 0 open
- Connection: full marks, safe for how it connects
Fix first: Upkeep is hard to verify: pin the version you tested.

If a later release changes any of this, we re-grade it the same way we re-grade every listing.

What a grade can't tell you

A grade comes from four public signals: login, upkeep, known CVEs and how the server connects. That's useful for a first pass. But it can't tell you:

What a server does when it runs. We don't read every line of code. A server with an A can still have a tool description that steers your agent wrong. See MCP prompt injection and MCP tool annotations.
Anything about servers we haven't graded. Local scripts, private servers and most registry entries come back as "not found" or "only in the official registry". Review those yourself, or ask us to grade one.
Whether today's grade still holds. We re-score after releases and new CVEs, but there's always a gap. Check the date on the listing.
How you set it up. A well-graded server with an all-access token, run over plain HTTP, is still a risk. The MCP security best practices cover that part.

Use the grade to decide where to look first, not as the final word.

Frequently asked questions

Is the TrustedMCP MCP server free?

Yes. The server, the npm package and the API key are free. The key needs an email address, and it allows 300 requests a minute.

Do I need an API key?

Yes, for tool calls, from version 0.2.0. Your client can connect and list the tools without one, but each call returns a message asking for a key. Get one at trustedmcp.io/developers.

Does it send my MCP config or my tokens to TrustedMCP?

The npx version doesn't. It sends server names, package names, image names and server URLs, and keeps tokens, env values, headers and file paths on your machine. The hosted version can't read your files, so you paste the config, and that text is sent over HTTPS. We read only the same fields from it.

Is the TrustedMCP MCP server open source?

Yes. The code is on GitHub under the MIT license, the package is trustedmcp on npm, and the server is in the official MCP Registry as io.trustedmcp/trustedmcp.

Why should I trust a grade you give your own server?

You don't have to. The rubric is public, the four checks are on the listing, and the code is open, so you can check each point yourself. We grade it with the same rubric as every other listing, and we keep it out of other listings' alternatives and our "best" lists.

How do I add TrustedMCP to Claude Code?

Run claude mcp add --transport http trustedmcp https://trustedmcp.io/mcp --header "Authorization: Bearer YOUR_KEY" for the hosted server, or claude mcp add trustedmcp -e TRUSTEDMCP_API_KEY=YOUR_KEY -- npx -y trustedmcp mcp to run it locally.

How do I check an MCP server before installing it?

Add the TrustedMCP MCP server, then ask your agent for the grade: "What grade does @playwright/mcp have?" runs lookup_server, and "Check the MCP servers in my config" runs check_config on all of them at once. For an official MCP Registry entry, registry_lookup shows its publisher, packages and reported issues. See also how to check an official MCP Registry entry and how to scan your MCP config file.

Does it replace an MCP security scanner?

No. A grade comes from public signals and doesn't read tool descriptions or run the server. MCP security tools covers what scanners, inspectors and gateways check.

Which servers can it grade?

Every server listed on trustedmcp.io/servers. For anything else, check_config and registry_lookup still show what the official MCP Registry knows. To get a server graded, submit it.