AgentDomains MCP server

MCP server for AgentDomains — free domains for the websites and APIs AI agents build. It gives any MCP-speaking agent a real hostname under makes.fyi or agentdomains.co, with DNS records, HTTPS, URL forwarding, and reverse proxying.

Two ways to run it: locally over stdio via npx, or against the hosted remote endpoint at https://mcp.agentdomains.co.

Install (stdio)

No install step — npx fetches it on demand:

AGENTDOMAINS_API_KEY=adom_… npx -y @agentdomains/mcp

Claude Code

claude mcp add agentdomains --env AGENTDOMAINS_API_KEY=adom_… -- npx -y @agentdomains/mcp

Claude Desktop and other config-file clients

Add to claude_desktop_config.json (or your client's equivalent):

{
  "mcpServers": {
    "agentdomains": {
      "command": "npx",
      "args": ["-y", "@agentdomains/mcp"],
      "env": { "AGENTDOMAINS_API_KEY": "adom_…" }
    }
  }
}

Related MCP server: domain-suite-mcp

Remote endpoint

https://mcp.agentdomains.co

Streamable HTTP; send JSON-RPC over POST with your key in the Authorization header:

curl -s https://mcp.agentdomains.co \
  -H 'Authorization: Bearer adom_…' \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

In Claude Code: claude mcp add --transport http agentdomains https://mcp.agentdomains.co --header "Authorization: Bearer adom_…"

Configuration

If you already use the Go CLI, its ~/.agentdomains/config.json is picked up automatically and no env var is needed.

No key yet? Call the signup tool — it needs no credentials and returns one. The key is shown once and is never retrievable, so store it immediately. Accounts are provisional: attach an email with attach_email and confirm it within 30 days or the account and its domains are deleted.

The API key is never logged, printed, or included in any tool output.

Tools

Every label-scoped tool takes an optional domain (makes.fyi, the default, or agentdomains.co). There is no endpoint for deleting an individual DNS record.

On HTTPS

"Get me SSL" is not one operation here. Two honest paths:

• set_proxy — the edge terminates TLS with its own certificate and proxies to your backend. Nothing to install on the origin.

• add_acme_challenge — publishes the DNS-01 TXT record so your own ACME client can issue a Let's Encrypt certificate (needed for wildcards or hosts with no public port 80). A plain public web server can just use HTTP-01 and needs no DNS record at all.

Compatibility

Any MCP client works. Verified shapes for the common ones:

• Claude Code — claude mcp add (stdio or --transport http), as above.

• Claude Desktop — mcpServers block in claude_desktop_config.json.

• OpenAI Agents SDK — MCPServerStdio(params={"command": "npx", "args": ["-y", "@agentdomains/mcp"]}), or MCPServerStreamableHttp against the remote URL.

• LangChain / LangGraph — langchain-mcp-adapters: MultiServerMCPClient({"agentdomains": {"command": "npx", "args": ["-y", "@agentdomains/mcp"], "transport": "stdio"}}).

• CrewAI — MCPServerAdapter with the same stdio parameters.

• Anything else — point your client at npx -y @agentdomains/mcp over stdio, or at https://mcp.agentdomains.co over Streamable HTTP.

Development

npm install
npm run build     # tsc -> dist/
npm run smoke     # boots dist/index.js, runs initialize + tools/list over stdio

Requires Node 18+. The only runtime dependency is @modelcontextprotocol/sdk.

Related

• CLI (Go) — the agentdomains command

• Agent skill — Claude Code plugin

• Docs · HTTP API

License

FSL-1.1-Apache-2.0 — see LICENSE. Same terms as the CLI.