M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
O

OmniRoute

Community
Gateway / proxy
A free AI gateway for 290+ providers through one endpoint

This gateway aggregates or proxies other MCP servers; it only becomes useful once you connect real servers behind it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #199 of 438
Stars
★ 78.1k Very popular
Transport
stdio (local process) · SSE · Streamable HTTP
Runtime
Node.js · Docker
Credentials
No credential needed
License
MIT
Last commit
54FMRS · D

The source presents OmniRoute as a broad open-source local AI gateway focused on multi-provider routing, free-tier aggregation, fallback resilience, and token compression. No individual MCP tool identifiers are provided, so the tools list is left empty.

Strongest · Documentation 13/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
8/20
Maintenance
13/20
Documentation
13/20
Setup experience
12/20
Why each score
Reliability 8/20
The README describes stdio, HTTP, and SSE MCP entry points, 104 tools, and JSON-RPC/A2A capabilities, with a connection example that makes the happy path plausible. However, the server manifest is unavailable, and no implementation, locked dependencies, CI workflow, or committed tests are supplied. Claims such as “25,000+ tests” cannot be verified, so points are deducted and the static calibration cap keeps the score at or below 12.
Security and permissions 8/20
The README claims local-first operation, no telemetry by default, AES-256-GCM credential encryption, scoped tokens, audit logs, loopback restrictions, and prompt-injection protection. These are positive signals. However, MCP is also described as controlling the whole gateway and accessing keys, providers, routing, and memory; permissions, confirmations, data flows, and defaults for remote mode, TLS/MITM decryption, and external provider traffic are not verifiable from the supplied source. No explicit red-line risk is established, but the evidence is insufficient for a high score.
Maintenance 13/20
The project has an MIT license, is not archived, and the README reports changes from v3.8.20 through v3.8.49, 500+ contributors, open issues, and community channels. This indicates activity and some ownership signals. Points are deducted because no verifiable commit history, release process, dependency-update policy, vulnerability-response policy, or explicit security contact is provided; popularity and star counts do not add points.
Documentation 13/20
The README covers installation, compatible clients, MCP endpoints, routing, compression, deployment platforms, authentication/remote tokens, and several architectural areas, with multiple command examples. Points are deducted because it is highly marketing-oriented, while the manifest, actual tool parameters and return values, error semantics, limits, permission matrix, cost details, and privacy boundaries lack reviewable source evidence. The referenced detailed documents were not supplied, so best-case assumptions are not justified.
Setup experience 12/20
The README provides npm, Docker, desktop, Termux, and source installation paths, plus `omniroute --mcp`, an HTTP MCP URL, and a Claude Code configuration example. Mainstream-client setup therefore appears reasonably clear. Points are deducted because verifiable runtime versions, dependency outcomes, first-start or health checks, authentication defaults, and troubleshooting are missing; the manifest is unavailable and there is no supplied CI or test evidence covering setup.

Static review · not runListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Developers who need a unified AI endpoint
  • Users of Claude Code, Cursor, Cline, Codex, or other coding agents
  • Teams interested in free tiers, cost-aware routing, and local deployment

Not for

  • Users who only need one model through its official client
  • Users unwilling to manage provider keys, quotas, and routing policies
  • Users seeking a hosted cloud service or official product integration not documented in the source

Required permissions

  • Access to the local OmniRoute gateway and its MCP interfaces
  • Use of configured provider credentials or built-in keyless free providers
  • MCP access to routing, providers, combos, cache, compression, and memory; remote tokens support read, write, and admin scopes
  • Process-spawning routes remain restricted to loopback access

Risks and side effects

  • Automatic fallback may send requests to different providers or models
  • Remote MCP exposes access to 104 tools, subject to token scopes and configuration
  • Free tiers, quotas, provider terms, and availability may change; the README notes ToS flags for some providers
  • Aggressive compression can alter context wording, although code, URLs, JSON, and structured data are preserved

Setup

Before you start

Runtime：Node.js · Docker

From the repository root, run npm install && npm run dev, or install globally with npm install -g omniroute. Start the stdio MCP server with omniroute --mcp, or connect to http://localhost:20128/api/mcp/stream or /api/mcp/sse. The README also provides this Claude Code command: claude mcp add-server omniroute --type http --url http://localhost:20128/api/mcp/stream.

Check that it works

After installing with npm i -g omniroute and starting the server, connect via MCP (e.g. claude mcp add-server omniroute --type http --url http://localhost:20128/api/mcp/stream) and confirm the 104 MCP tools appear in the client's tool list, or curl http://localhost:20128/v1/chat/completions with model auto to confirm a keyless reply.

Troubleshooting

  1. Confirm that the local service is running at http://localhost:20128
  2. Run omniroute doctor to inspect providers, ports, and native dependencies
  3. Verify the MCP endpoint path: /api/mcp/stream or /api/mcp/sse
  4. If auto does not respond, check the health and quota status of built-in or configured providers
  5. For remote mode, verify the selected context, token scope, and connection settings

Things to try

Once connected, you can ask your AI assistant things like:

  • Send a message using the auto model to test the zero-config free routing through the gateway
  • List the connected providers and check their health status
  • Configure my Codex CLI to use an OmniRoute remote model
  • Show how many free-tier tokens are remaining this month

Use cases

Unifying access to multiple AI providers and models
Configuring automatic routing and quota fallback for coding agents
Reducing tool-output and context token usage through compression
Allowing MCP or A2A agents to manage gateway providers, combos, caching, and memory

Supported clients

Claude Desktop
Cursor
Claude Code

Listed from the project's documentation, not tested by this site.

Overview

OmniRoute is an MIT-licensed, local-first AI gateway that connects one endpoint to 290+ providers and 500+ models, including 90+ providers with free tiers. It offers quota-aware automatic fallback, 19 routing strategies, RTK and Caveman token compression, MCP and A2A support, and compatibility with multiple coding tools.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 40523566e611 Data synced 2026-10-19 Read the FMRS scoring method