Skip to main content

Integrations

Agent Router presents an OpenAI-compatible surface, so most tools that already speak to an AI provider connect by changing only a base URL and an API key. This page is the map of everything that plugs into the gateway: the SDKs and frameworks that send inference traffic, the coding agents that generate against it, the guardrail vendors and observability backends it forwards to, and the MCP clients that consume aggregated tools. Each entry links to the guide or reference that documents the connection.

SDKs and frameworks​

Any client that targets the OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages format works against the gateway. The base URL points at the proxy endpoint and the API key is an Agent Router key; no provider-specific client is required.

Agent Router SDK

Language quickstarts for Go, Python, TypeScript, and curl.

Integrate the gateway with an app

Point OpenAI, Anthropic, LangChain, and other SDKs at the gateway.

The current documentation URLs for the supported SDKs and frameworks (OpenAI, Anthropic, LangChain, Vercel AI SDK, Pydantic AI, CrewAI, and others) are listed in External links.

Coding agents​

Coding agents receive a ready-to-paste config that teaches them the gateway base URL, authentication, and endpoints, so generated code routes through Agent Router from the first request.

Coding agents & tools

Config for Claude Code, Cursor, Codex CLI, and Lovable.

Playground

Compare models and iterate on prompts before wiring an agent.

Guardrail vendors​

Guardrails run inline in the gateway and can call out to external content-safety services for evaluation. Vendor and custom guardrails are configured by a platform operator.

Configure vendor guardrails

Connect an external content-safety service as a guardrail provider.

Configure custom guardrails

Define regex, keyword, and PII rules enforced inline by the gateway.

Observability backends​

The gateway exports traces and metrics over OpenTelemetry (OTLP) and exposes a Prometheus scrape endpoint, so any OTLP-compatible or Prometheus-compatible backend can receive gateway telemetry.

Export telemetry to an observability stack

Send traces and metrics to Grafana, Datadog, New Relic, and similar backends.

OpenTelemetry metrics

The metric names and attributes the gateway emits.

MCP clients​

Multiple MCP servers can be aggregated into a single profile and exposed to AI clients through one endpoint, with access governed centrally.

Aggregate MCP servers into a profile

Combine MCP servers into one endpoint for AI clients.

Govern MCP server access

Control which MCP servers and tools are reachable, and by whom.