Skip to Content
Server Tools

Server Tools

Server tools run on our side of the wire: the gateway injects the tool, executes it when the model calls it, feeds the result back to the model, and loops until the model finishes. You get the final answer in one request — no client-side tool loop.

ToolBacked byWhat it does
Web searchFirecrawlSearches the web and returns result snippets + URLs to the model
Web fetchFirecrawlFetches a specific URL and returns extracted page content
PDF parsingMistral OCRExtracts text (incl. scans) from PDF/file inputs before the model sees them

Activating tools

Activation follows your ingress format’s idiom, at functional parity with OpenRouter’s server-tools suite.

Attach the web plugin to any request (OpenAI-format shown; the same plugins extension field applies to all three ingress formats):

{
  "model": "anthropic/claude-sonnet-4.5",
  "messages": [
    { "role": "user", "content": "What changed in the EU AI Act this month?" }
  ],
  "plugins": [{ "id": "web" }]
}

The model decides when to search or fetch; tool activity is visible in the response’s tool-call trace and itemized in your usage events.

Pricing

Server-tool executions are billed against your workspace credits as separate, itemized ledger lines tied to the same request — your activity page and monthly statements show model usage and tool usage side by side, per request. Each tool’s per-use rate is published on the Pricing page.

The server-tools API surface is the newest part of the gateway; field names follow the OpenRouter plugin idiom and will be frozen with the generated OpenAPI schema. Check this page and the changelog before hard-coding activation payloads.

Models & RoutingErrors