---
title: "OpenRouter MCP Server: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/openrouter/
description: "OpenRouter-hosted MCP server for live model search, per-provider pricing, credits, usage rankings and docs, plus two billable tools (send-message and generate-image). Connecting mints a dedicated OpenRouter key with a 7-day expiry and a $10 default spend limit."
---

# OpenRouter MCP Server: review, setup & limitations · Undominated.ai

> OpenRouter-hosted MCP server for live model search, per-provider pricing, credits, usage rankings and docs, plus two billable tools (send-message and generate-image). Connecting mints a dedicated OpenRouter key with a 7-day expiry and a $10 default spend limit.

[← Explore all mcp servers](/mcp-servers/)

MODEL GATEWAY AND PRICING / OpenRouter

# OpenRouter MCP Server

OpenRouter-hosted MCP server for live model search, per-provider pricing, credits, usage rankings and docs, plus two billable tools (send-message and generate-image). Connecting mints a dedicated OpenRouter key with a 7-day expiry and a $10 default spend limit.

 [See setup guidance ↓](#setup)[Original source ↗](https://openrouter.ai/)

SOURCE REVIEW

 Reviewed 2026-10-13
 Evidence 2 linked sources
 Publisher OpenRouter
 Licence Not established
 [Read what was—and wasn’t—checked ↓](#review)

“The OpenRouter MCP server plugs OpenRouter into the AI tools you already use.”

 [OpenRouter · upstream description ↗](https://openrouter.ai/docs/guides/overview/mcp-server) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - Publisher-hosted remote endpoint needs no local install and covers model discovery plus live per-provider pricing in one place.
- The guide names the billable tools (send-message, generate-image) and the one that writes (send-feedback), and caps each connection with its own key: 7-day expiry and a $10 default spend limit.

### A good fit for

 - Looking up which models OpenRouter serves, which providers serve them and at what price, before choosing one.
- Sending a test prompt or image request to a model from the editor to compare models, billed to the account.

### Weigh up before choosing

 - send-message and generate-image make billable inference calls on the account; send-feedback writes feedback on one of your own generations. The other tools are read-only lookups.
- Its list-benchmarks and list-models tools surface Artificial Analysis and Design Arena scores.
- The guide positions it as a build-time aid: to run models in an application, call the OpenRouter API directly.
- Every request needs the OAuth-issued key: an unauthenticated request returns 401, and the key expires after 7 days, after which the client must authenticate again.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - Hosted MCP guide, read 2026-10-13 (HTML and its Markdown export): endpoint https://mcp.openrouter.ai/mcp, per-client setup, the tool table, the OAuth/PKCE flow and the dedicated key with a 7-day expiry and $10 default spend limit.
- OAuth authorization-server metadata served by the endpoint on 2026-10-13: authorization, token and dynamic registration endpoints, code grant only, PKCE S256. An unauthenticated POST to https://mcp.openrouter.ai/mcp answered 401 with a WWW-Authenticate header pointing to the protected-resource metadata.

### Our findings

 - The server is remote-only; there is no repository, package or revision to pin.
- The guide's tool table lists send-message, generate-image, list-models, get-model, list-model-endpoints, list-providers, list-daily-model-rankings, list-app-rankings, get-credits, get-generation, list-benchmarks, list-task-classifications, search-docs, send-feedback, spawn-ori-eval, install-ori-harness and ping.
- The issued token is a standard OpenRouter API key labelled "OpenRouter MCP: <app name>", visible and revocable in the key dashboard; its spend cap can be edited on the approval screen.
- The transport is recorded as Streamable HTTP from the guide's claude mcp add --transport http command; the guide does not name the transport itself.
- configExample is the guide's Cursor CLI ~/.cursor/mcp.json block.
- Corrected 2026-10-13: the upstream description was not a quote from the guide; the record also said catalogue and pricing lookups work without an account and that the client attaches an API key, while the guide says every request needs the OAuth-issued key.

### Not established by this review

 - Server connection, authentication and tool execution were not tested.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://openrouter.ai/docs/guides/overview/mcp-server)
 - Add https://mcp.openrouter.ai/mcp to the MCP client (Claude Code: claude mcp add --transport http openrouter https://mcp.openrouter.ai/mcp).
- Run the client's login step and approve the OpenRouter consent page, which mints a dedicated key; adjust its spend cap there if $10 is not wanted.
- Confirm which tools are billable before enabling send-message or generate-image.

### Before you start

 - An OpenRouter account.

### Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

 {
 "mcpServers": {
 "openrouter": {
 "url": "https://mcp.openrouter.ai/mcp"
 }
 }
} Copy configuration ↗

### Compatibility

MCP clients that add a remote server by URL and run OAuth: the guide gives setup for Claude Code, Codex CLI, OpenCode, Cursor CLI and Claude Desktop / web custom connectors. · Requires an OpenRouter account; the connection key is minted at the consent page.

### Transports

Streamable HTTP

### OpenRouter OAuth (dedicated per-connection key)

 - Every tool call uses a dedicated OpenRouter API key minted by OAuth with PKCE, expiring after 7 days with a $10 default spend limit.
- send-message and generate-image spend account credit; send-feedback sends feedback on one of your own generations to OpenRouter.
- The guide says no source code is sent anywhere unless send-message, generate-image or send-feedback is called.

### Cost model

send-message and generate-image are billable inference calls on the OpenRouter account, bounded by the connection key's spend limit ($10 by default). The guide marks the other tools as read-only lookups and names no charge for them.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [OpenRouter hosted MCP guide ↗](https://openrouter.ai/docs/guides/overview/mcp-server) Checked 2026-10-13 https://openrouter.ai/docs/guides/overview/mcp-server Supports: summary, upstreamDescription, install, compatibility, access, transports, configExample, limitations, whySelected, bestFor, review
- [OpenRouter MCP OAuth authorization-server metadata ↗](https://mcp.openrouter.ai/.well-known/oauth-authorization-server) Checked 2026-10-13 https://mcp.openrouter.ai/.well-known/oauth-authorization-server Supports: access, review

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### AWS Pricing MCP Server ↗ Reads AWS price-list data through a first-party local server instead of asking a model to recall a rate.](/mcp-servers/awslabs-aws-pricing/)[### Hugging Face MCP ↗ Connects an assistant to Hub discovery, with optional community tools and account-scoped compute or repository actions.](/mcp-servers/hugging-face/)[### Hugging Face Official MCP Server ↗ Official Hugging Face MCP server connecting AI assistants to the Hub, repository search, and Gradio AI applications.](/mcp-servers/huggingface/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
