🤖 MCP Setup · 60 seconds

Connect OmniDeploy to Claude Desktop

OmniDeploy is the first AI inference router with native Model Context Protocol support. Add it as a tool inside Claude Desktop, Cursor, Cline, or any MCP client and your agent can route inference, query pricing, and list providers — programmatically.

1

Install mcp-proxy

OmniDeploy speaks MCP over HTTP/JSON-RPC 2.0. Most MCP clients today use stdio, so we pipe through mcp-proxy (open source).

terminalbash
# pick one
npm install -g mcp-proxy
# or
pip install mcp-proxy
2

Edit your Claude Desktop config

On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%\Claude\claude_desktop_config.json

claude_desktop_config.jsonjson
{
  "mcpServers": {
    "omnideploy": {
      "command": "mcp-proxy",
      "args": [
        "--transport", "streamablehttp",
        "https://omnideployservice.online/mcp"
      ],
      "env": {
        "MCP_AUTH_HEADER": "X-API-Key: <YOUR_OMNIDEPLOY_KEY>"
      }
    }
  }
}

⚠ Replace <YOUR_OMNIDEPLOY_KEY> with your real OmniDeploy key (starts with omni_live_). Don't have one? Get one in 30 seconds here →

3

Restart Claude Desktop and try it

Quit Claude Desktop completely, reopen it. In a new chat, ask:

Try in Claude

“What OmniDeploy tools do you have? Use route_inference to ask llama-3.1-8b-instant for a haiku about Mumbai.”

⚡

Or test the MCP endpoint directly

You don't need an MCP client at all — the server speaks plain JSON-RPC 2.0 over HTTP:

terminalbash
# Discover what tools OmniDeploy exposes
curl -X POST https://omnideployservice.online/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# Call route_inference (auth required for actual inference)
curl -X POST https://omnideployservice.online/mcp \
  -H "Content-Type: application/json" \
  -H "X-API-Key: <YOUR_OMNIDEPLOY_KEY>" \
  -d '{
    "jsonrpc":"2.0","id":2,"method":"tools/call",
    "params":{
      "name":"route_inference",
      "arguments":{
        "model":"llama-3.1-8b-instant",
        "messages":[{"role":"user","content":"Hello"}]
      }
    }
  }'
✨

Cursor, Cline, Continue, Zed

All MCP-compatible clients accept the same JSON config. In Cursor: Settings → Model Context Protocol → Add server. In Cline (VS Code): edit cline_mcp_settings.json.

What your agent can do once connected

route_inference

Send a chat-style prompt. We pick the cheapest provider that meets your policy and return cost + latency in metadata.

get_pricing_snapshot

Query live per-token pricing across all 13 providers. Filter by model name. Useful for cost-aware agents.

list_providers

Enumerate every provider OmniDeploy can route to, plus their supported models.

Works in any MCP client

Same JSON-RPC 2.0 endpoint, same three tools. Works in:

  • Claude Desktop
  • Cursor
  • Cline (VS Code)
  • Continue.dev
  • Zed AI
  • Custom agents

Don't have an OmniDeploy API key yet?

Provision one in 30 seconds — email only, no password.

Get an instant API key →Sign up for full account