Give your agent the internet in one minute.

Paste one prompt into Claude Code, Cursor, Codex or any MCP client. Your agent adds the server, asks to join your workspace and waits for you to approve it.

Set up by hand

Paste the prompt
Your agent adds the MCP server and calls start_login. No API key goes through the chat.
Approve it
Open the link it shows you and check the code. The free plan includes 3 agents.
Ask your agent
It discovers the tools by itself. Try a prompt from the list below.

Or set it up by hand

Run in your terminal

claude mcp add --transport http socket-help https://socket.help/mcp \
  --header "Authorization: Bearer $SOCKET_HELP_API_KEY"

Leave out the --header line to try it without an account; anonymous limits apply.

Prefer a terminal or code? The same tools ship as npm install --global @socket-help/cli@beta and npm install @socket-help/sdk@beta. See the CLI and SDK docs.

Try one of these prompts

Create a socket.help webhook inbox, wait for the first request, and explain what it contains.
Create a socket.help email inbox, sign up at my staging site with it, and finish verification.
Before you deploy, use socket.help ask_human to get my approval and wait for my answer.
Mock GET /users/:id with three realistic users, then call it to check the response.
Look up the DNS records and SSL certificate for example.com and summarize any problems.

What your agent can do

Tools appear automatically after you connect. Utilities stay behind search_tools so your agent’s tool list stays short.

socket.help MCP tools
Tool	What it does
`create_webhook_inbox`	A private URL that captures any HTTP request.
`wait_for_webhook`	Block until the inbox receives a request.
`inspect_webhook_inbox`	Read captured requests: method, headers, body.
`create_email_inbox`	A receive-only address for sign-up and reset tests.
`wait_for_email`	Block until mail arrives; codes and links come extracted.
`create_mock_api`	Routes, templates, CRUD records and failure modes.
`call_mock_api`	Call a mock or change its CRUD data.
`upload_temporary_file`	Share a file through a private, expiring link.
`memory_set / memory_get`	Remember JSON between sessions (needs an API key).
`ask_human`	Ask you before a risky step; you answer from a link.
`wait_for_approval`	Block until you answer, then act on it.
`create_monitor`	Watch a URL, a certificate or a cron job, with alerts.
`search_tools`	Find one of 130+ utilities by keyword.
`run_tool`	Run a utility: JSON, JWT, DNS, SSL, PDF, images and more.
`resource_request`	List, renew, share or delete what your agent created.

Limits and safety

Private by default

Everything your agent creates is private, with separate tokens to use and to manage it. Review destructive tool calls in your client.

Same limits everywhere

MCP, the API, the CLI and the browser share your plan’s allowances. See Plans for the numbers.