MCP server

One MCP server for every tool your agent needs

Q: How many requests can it make?

600 requests a minute per key to /mcp, within your plan's shared request limit.

Connect Claude Code, Cursor or any MCP client to Arettic, and your agent can search the web, read pages, find contacts and pull company data through one server. Failed calls are free.

Arettic is a hosted MCP server. Add it once with one agent key, and your agent gets two ways to buy. Guaranteed jobs are charged only when the result passes a published check. Provider endpoints are called the way each provider documents them, and cost nothing when the provider fails.

Connect in one step

The server speaks Streamable HTTP at https://api.arettic.com/mcp. Send your agent key as a Bearer token.

Claude Code

claude mcp add --transport http arettic https://api.arettic.com/mcp --header "Authorization: Bearer sk_test_…"

Cursor and other clients (mcp.json)

{
  "mcpServers": {
    "arettic": {
      "url": "https://api.arettic.com/mcp",
      "headers": {
        "Authorization": "Bearer sk_test_…"
      }
    }
  }
}

Test keys (sk_test_…) are free and use mock providers. The @arettic/mcp bridge, for clients that only run local servers, isn't on npm yet. Until it is, use the hosted endpoint. Every client config is in the MCP server docs.

What your agent can do with it

The eleven MCP tools
Tool	What it does
`recommend`	Ranks the tools for a job by score, then price. Free.
`execute`	Runs a tool. Charged only if the result passes its check.
`search_endpoints`	Finds provider endpoints: search, scraping, contacts and more.
`call_endpoint`	Calls one with the provider's own parameters. Free if it fails.
`list_task_types`	The guaranteed jobs, each with its pass rule.
`get_tool`	One tool's scores, history and schemas.
`get_job`	Progress and results of a batch of more than 25 items.
`get_receipt`	What a purchase cost, how it was checked, and any refund.
`open_dispute`	Dispute a charged result within 7 days.
`get_approval`	Whether a person has approved a big purchase yet.
`get_balance`	The agent's budget and the org's credits.

A typical session

recommend with the job, for example find_verified_email in the US. Free.
execute with the first tool_id marked purchasable and the input. Add fallback: true to try the next-ranked tool once if an item fails.
The result comes back passed or failed. A pass is charged. A fail is refunded, and your agent gets the reason.
For anything else, search_endpoints finds a provider endpoint and call_endpoint calls it.

What it costs

Connecting is free, and so are recommend, list_task_types and get_tool.
execute charges the job's price only for results that pass.
call_endpoint charges per call. Nothing is charged if the provider errors or times out.
Everything counts toward the agent's monthly budget, so it can't run up a bill. See spending limits.
New accounts get $1 of free credit. Prices are on Prices.

Why one server, not one per provider

Many providers ship their own MCP server. Each needs its own account, key and bill, and none tells your agent which provider is best for the job. One Arettic server puts every provider on one key and one balance, ranks tools by public score, and refunds results that fail.

Questions

Is the Arettic MCP server free?

Connecting is free, and so are recommendations and the catalogue. You pay for results that pass and for endpoint calls that work. New accounts get $1 of free credit.

Which MCP clients does it work with?

Claude Code, Cursor and any client that supports Streamable HTTP with a custom header. Clients that only run local servers, like Claude Desktop, can use the bridge once @arettic/mcp is on npm.

Do I need an account with each provider?

No. Your Arettic agent key covers every provider in the catalogue, with one balance and one bill.

Can my agent overspend through MCP?

No. Each agent key has a monthly budget ($50 by default) and an approval threshold ($20 by default). Over either, the purchase waits for a person.

How many requests can it make?

600 requests a minute per key to /mcp, within your plan's shared request limit.

Updated 2026-10-23 · This page as Markdown · JSON