Remote MCP server
Multi-model consensus inside Claude — one URL, no install
Tokonomix runs a live remote MCP server at https://mcp.tokonomix.ai/mcp. Add it to claude.ai as a custom connector and Claude gains eight tools: cross-check an answer across Claude, GPT and Gemini with an independent judge, query the live model catalog with EU routing, and feed ratings back into model reputation. Billing runs on your own Tokonomix account.
Connect in four steps
- Open the connector settings. In claude.ai go to Settings → Connectors → Add custom connector.
- Paste the server URL. Enter https://mcp.tokonomix.ai/mcp as the remote MCP server URL. No client ID, secret or token is needed.
- Connect and log in. Click Connect. You are redirected to tokonomix.ai — log in with the Tokonomix account that should be billed for usage.
- Approve the scopes. Approve the consent screen. The Tokonomix tools appear in your Claude chats immediately.
The eight tools
| Tool | Scope | What it does |
|---|---|---|
| tokonomix_consensus_ask | consensus:run | Ask Claude + GPT + Gemini side by side; an independent judge synthesizes one verdict with per-model agreement. |
| tokonomix_single_ask | consensus:run | Cheap single-model passthrough to any model in the catalog. |
| tokonomix_list_models | consensus:read | Live model catalog with prices, speed and EU-routing options. |
| tokonomix_get_balance | consensus:read | Current credit balance of the connected account. |
| tokonomix_get_skill | consensus:read | The usage skill that teaches an agent when and how to use the council. |
| tokonomix_skill_version | consensus:read | Version/fingerprint of the skill, for pinning and reproducibility. |
| tokonomix_rate_consensus | feedback:write | Rate a consensus answer 1–10; feeds model reputation and council selection. |
| tokonomix_relay_human_feedback | feedback:write | Relay structured human feedback on a council round (earns credit rewards where enabled). |
Billing and limits, stated plainly
- Usage is billed to the Tokonomix account you logged in with — same per-model prices as the API, no connector surcharge. No balance → tool calls return 402.
- The remote transport does not support
mode:"full", vision/image input, or the onboarding/upload tools — those live in the npm/local transport. - Scopes are enforced server-side per tool; the consent screen names all of them up front so you authorize once.
FAQ
- What is a remote MCP server?
- MCP (Model Context Protocol) is the open standard that lets AI assistants use external tools. A remote MCP server is one you connect by URL — no package install, no local process. Claude talks to it over HTTPS with OAuth login.
- Do I need an API key or client ID?
- No. The connector uses OAuth: you paste the URL, log in with your Tokonomix account and approve the scopes. Keys and secrets never touch your Claude configuration.
- What does it cost?
- Tool calls are billed to the Tokonomix account you log in with, at the same per-model prices as the API — no connector surcharge. If the account has no balance, tool calls return a 402 error; that is an empty wallet, not an outage.
- How is this different from the npm package?
- The npm transport (npx tokonomix-council-mcp) runs locally with an API key and supports a few extras the remote transport does not: mode "full", vision/image input, and the onboarding/upload tools. The remote transport is the fastest way in — one URL, no install.
Part of a bigger agent stack
Tokonomix is the consensus and metering layer of an open agent ecosystem. Related building blocks: agentixmesh — agent trust layer for safe agent-to-agent messaging, AgentsWeaver — agent job orchestration & metering for distributed agent jobs, and Agents Never Sleep for unattended runs. See also all integrations and the ecosystem interop page.
Last updated: 2026-07-27 · Server: https://mcp.tokonomix.ai/mcp · Status: live