MCP server.

The server lives at https://metasearch.sh/mcp. It speaks streamable HTTP, stateless, on protocol 2026-07-28 and the 2025 revisions before it, and every request carries your API key as a bearer token. Tool calls go through the same validation, rate limits and ledger as POST /api/v1/search, so a search costs the same here as anywhere else.

Get a key

Sign in, then create a key in the dashboard. New accounts start with $1.00 of credit. Put the key in your shell profile so every client below can read it:

$ export METASEARCH_API_KEY=ms_live_…

Claude Code

$ claude mcp add --transport http metasearch https://metasearch.sh/mcp \
  --header "Authorization: Bearer $METASEARCH_API_KEY"

Add --scope user to use it in every project, or --scope project to share it through .mcp.json. Check it with claude mcp list.

Cursor

{
  "mcpServers": {
    "metasearch": {
      "url": "https://metasearch.sh/mcp",
      "headers": { "Authorization": "Bearer ${env:METASEARCH_API_KEY}" }
    }
  }
}

VS Code

Install in VS Code and paste your key when asked, or add it to the workspace by hand:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "metasearch-api-key",
      "description": "metasearch.sh API key",
      "password": true
    }
  ],
  "servers": {
    "metasearch": {
      "type": "http",
      "url": "https://metasearch.sh/mcp",
      "headers": {
        "Authorization": "Bearer ${input:metasearch-api-key}"
      }
    }
  }
}

Codex

$ codex mcp add metasearch --url https://metasearch.sh/mcp --bearer-token-env-var METASEARCH_API_KEY

[mcp_servers.metasearch]
url = "https://metasearch.sh/mcp"
bearer_token_env_var = "METASEARCH_API_KEY"

Other clients

Any client that supports remote MCP servers over HTTP works: give it the URL https://metasearch.sh/mcp and the header Authorization: Bearer ms_live_…. A request with a missing, unknown or revoked key gets HTTP 401, so the client can ask for a new key. If the key cannot be checked during an outage, the answer is HTTP 503 with a Retry-After header.

Tools

Leave mode out and the server picks one from the query, then says which and why in routing.auto; see Routing. It never picks deep, and picks social only when the query names X or Twitter, because those cost several times more: ask for them by name. One entry in providers pins that provider; several are queried together and their results merged.

Each tool returns a short text list for the model and the full API response as structured content. The text ends with one line of billing:

Mode news (auto, jev): Asked about recent events (p 0.97); used news with freshness week.
1. Solid-state batteries clear a manufacturing hurdle
   https://…
   Toyota and Idemitsu say their sulfide electrolyte line…
2. …
Charged $0.0015 · balance $0.9985 · served by serper

Pages from fetch start with a line marking them as untrusted web content. Treat that text as data: a page can contain instructions aimed at your agent.

What it costs

Tools are priced exactly like the API: the provider's price for the call times 1.5, taken from your prepaid balance. A search starts at $0.00075 and a page fetch at $0.000375; empty and failed attempts cost nothing. Every provider and mode is on the pricing page.