Get started

Overview

ToolRouter is a unified gateway and marketplace for AI tools. The product goal is simple: agents should not need a different integration, billing surface, or discovery flow for every tool provider.

What is ToolRouter?

ToolRouter is a unified gateway and marketplace for AI tools. It normalizes authentication, billing, and response formats across every tool provider into a single interface, so agents and developers connect once and access the entire catalog through one API key.

Every tool provider has its own auth, billing, and response format. ToolRouter normalizes all of it into one interface.

  • A tool is the top-level product surface, such as SEO analysis or web search.
  • A skill is a callable operation within a tool, such as analyze_page or search. Each skill has its own input schema and examples.
  • A provider is an upstream API service (e.g. fal.ai, ElevenLabs) that executes skills. Tools can be served by multiple providers with automatic failover.
  • A caller authenticates once with an API key and can then discover and execute the whole catalog.

How do I connect to ToolRouter?

ToolRouter exposes the same catalog through four entrypoints: CLI for local inspection and execution, MCP for agent-native discovery via tools/list and tools/call, REST for JSON endpoints, and the web interface for human-readable docs and browsing.

ToolRouter exposes the same catalog through multiple entrypoints:

  • CLI: local-first inspection, execution, validation, testing, and plugin management.
  • MCP: agent-native discovery and execution through tools/list and tools/call.
  • REST: JSON endpoints for discovery, execution, key management, usage, and assets.
  • Web: human-readable docs, browse pages, rankings, and onboarding flows.

What tools are available?

The registry ships tools across categories including SEO, marketing, security, media, data, search, finance, and more. Each tool exposes one or more callable skills. Browse the full catalog at toolrouter.com/tools or run toolrouter tools in the CLI.

Some tools require upstream credentials (e.g. Firecrawl, Serper, ElevenLabs). Run toolrouter credentials to see which ones need setup.

What does a tool call return?

Every tool call returns a standardized ToolCallResult with four fields: status (success or failure), output (the skill-specific payload), usage (billing and cost metadata), and meta (tool identity and latency). This consistent response contract means agents can reason about any tool the same way.

Every call resolves to a standardized ToolCallResult. That matters because agents can reason about one response contract across the whole catalog:

  • status tells the caller if the call succeeded.
  • output contains the skill-specific payload.
  • usage contains billing and cost metadata.
  • meta contains tool identity and latency.

How do I use my own API keys?

You can pass your own provider API keys per-request via X-Provider-Key-* headers, store them in CLI config with toolrouter providers add, or set environment variable fallbacks. When you use your own keys, ToolRouter charges only a 5% platform fee instead of the full markup.

Tool authors can declare credential requirements in the manifest. ToolRouter then surfaces those requirements consistently in CLI and API discovery:

  • global provider keys can be stored in CLI config
  • environment fallbacks can satisfy credentials in local development
  • per-request headers can override provider keys for a single API call

Why should agents use ToolRouter?

Agents don't have to memorize a different auth scheme or response shape for every provider. They search one catalog, inspect one manifest format, call skills with JSON Schema input, and receive a consistent result envelope — letting them branch on usage, latency, and error type predictably.

Agents do not have to memorize a different auth scheme or response shape for every provider. They can:

  • search the catalog
  • inspect one manifest format
  • call skills with JSON Schema input
  • receive a consistent result envelope
  • branch on usage, latency, and error type in a predictable way

Where to go next

  • Quickstart if you want the shortest path to a working call.
  • Integration if you are wiring ToolRouter into an agent or backend.
  • API Reference for the complete REST endpoint documentation.
  • Billing for credits, pricing, and BYOK details.
  • Providers for upstream API routing, failover, and backend preferences.
  • Building Tools if you are extending the catalog.