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.
No install step — npx fetches it on demand:
AGENTDOMAINS_API_KEY=adom_… npx -y agentdomains-mcpclaude mcp add agentdomains --env AGENTDOMAINS_API_KEY=adom_… -- npx -y agentdomains-mcpAdd to claude_desktop_config.json (or your client's equivalent):
{
"mcpServers": {
"agentdomains": {
"command": "npx",
"args": ["-y", "agentdomains-mcp"],
"env": { "AGENTDOMAINS_API_KEY": "adom_…" }
}
}
}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_…"
| Variable | Purpose |
|---|---|
AGENTDOMAINS_API_KEY |
Your API key (adom_…). Falls back to api_key in ~/.agentdomains/config.json. |
AGENTDOMAINS_API_URL |
Override the API base URL. Defaults to https://api.agentdomains.co. |
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.
| Tool | Endpoint | Description |
|---|---|---|
check_availability |
GET /v1/available |
Is a label free? No key required. |
signup |
POST /v1/signup |
Create an account, get an API key (shown once). No key required. |
whoami |
GET /v1/whoami |
Account state, email, quota, usage. |
attach_email |
POST /v1/account/email |
Attach an email and send its verification link. |
claim_domain |
POST /v1/subdomains |
Register label.domain, optionally with a first record. |
list_domains |
GET /v1/subdomains |
List every name on the account. |
get_domain |
GET /v1/subdomains/{label} |
One name, its records, forward/proxy state. |
delete_domain |
DELETE /v1/subdomains/{label} |
Destructive — permanently delete a name and its records. |
add_dns_record |
POST /v1/subdomains/{label}/records |
Add an A, AAAA, CNAME, or TXT record. |
delete_record |
DELETE /v1/subdomains/{label}/records/{id} |
Destructive — remove one record, keeping the name. Ids come from get_domain. |
add_acme_challenge |
POST /v1/subdomains/{label}/records |
TXT at _acme-challenge for Let's Encrypt DNS-01. |
set_forward |
PUT /v1/subdomains/{label}/forward |
Redirect the name to a URL; replaces the address records on it. |
remove_forward |
DELETE /v1/subdomains/{label}/forward |
Stop forwarding. |
set_proxy |
PUT /v1/subdomains/{label}/proxy |
Serve a backend through the edge, with TLS terminated for you; replaces the address records on it. |
remove_proxy |
DELETE /v1/subdomains/{label}/proxy |
Stop proxying. |
delegate_nameservers |
PUT /v1/subdomains/{label}/ns |
Delegate the name to your own nameservers. |
delete_account |
DELETE /v1/account[?force=true] |
Destructive — close the account and kill its key; refuses while names are held unless force. |
upgrade_to_pro |
POST /v1/billing/checkout |
A Stripe checkout link for Pro (100 names, $5/month or $48/year). Charges nothing; a human opens the link and pays. |
manage_billing |
POST /v1/billing/portal |
A Stripe billing-portal link for a Pro account (card, invoices, interval, cancel). |
Every label-scoped tool takes an optional domain (makes.fyi, the default, or
agentdomains.co).
Setting a forward or a proxy takes the hostname over. The A/AAAA/CNAME records
on the label itself are deleted in the same call and handed back as replaced_records,
so tell the user what moved. Sub-label records (www.myapp.makes.fyi) and TXT records
are left alone, and if the forward fails to come up the replaced records are restored
with new ids.
A claim and its first record stand or fall together. If claim_domain is given a
record it cannot create, nothing is claimed — fix the record and call again. Re-claiming
a name the account already holds answers 409 with owned:true, which means carry on.
"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.
Any MCP client works. Verified shapes for the common ones:
- Claude Code —
claude mcp add(stdio or--transport http), as above. - Claude Desktop —
mcpServersblock inclaude_desktop_config.json. - OpenAI Agents SDK —
MCPServerStdio(params={"command": "npx", "args": ["-y", "agentdomains-mcp"]}), orMCPServerStreamableHttpagainst the remote URL. - LangChain / LangGraph —
langchain-mcp-adapters:MultiServerMCPClient({"agentdomains": {"command": "npx", "args": ["-y", "agentdomains-mcp"], "transport": "stdio"}}). - CrewAI —
MCPServerAdapterwith the same stdio parameters. - Anything else — point your client at
npx -y agentdomains-mcpover stdio, or athttps://mcp.agentdomains.coover Streamable HTTP.
npm install
npm run build # tsc -> dist/
npm run smoke # boots dist/index.js, runs initialize + tools/list over stdioRequires Node 18+. The only runtime dependency is @modelcontextprotocol/sdk.
- CLI (Go) — the
agentdomainscommand - Agent skill — Claude Code plugin
- Docs · HTTP API
FSL-1.1-Apache-2.0 — see LICENSE. Same terms as the CLI.
This server is listed in the official MCP Registry
as io.github.tashfeenahmed/agentdomains-mcp — the index PulseMCP, Glama and other
agent-tool directories read from. The metadata is server.json in this repo.
Publishing is manual (GitHub Actions is not running on this account). On each npm
release, bump version in server.json and its packages[0].version to match
package.json, publish to npm first, then:
# once: install the publisher (macOS/Linux)
curl -L "https://github.com/modelcontextprotocol/registry/releases/latest/download/mcp-publisher_$(uname -s | tr '[:upper:]' '[:lower:]')_$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/').tar.gz" | tar xz mcp-publisher
./mcp-publisher login github # device-code flow as the tashfeenahmed GitHub account
./mcp-publisher publish # uploads server.json
curl -s "https://registry.modelcontextprotocol.io/v0/servers?search=io.github.tashfeenahmed/agentdomains-mcp"The registry checks that the npm package's mcpName matches name in server.json,
so the npm version named in server.json must already be published.
docs/publish-mcp-registry.yml is an optional GitHub Actions workflow that does the same
on a v* tag via GitHub OIDC. It only runs if moved to .github/workflows/ and Actions
is available on the account.