Skip to content
Cursor logo (Anysphere)

Cursor via Tokonomix

Use Tokonomix models in Cursor's chat and add cross-family consensus to the Agent through MCP — one key, 132 models, region- and origin-routable.

Custom model + base URLMCP (Agent)AI-native IDEmacOS · Windows · Linux
Cursor — the AI code editor (Anysphere)

What is Cursor?

Cursor is an AI-native code editor from Anysphere — a fork of VS Code with three AI surfaces: Chat (ask about your code), the Agent / Composer (multi-file autonomous edits), and Tab (inline autocomplete). It's a desktop app for macOS, Windows and Linux.

Cursor lets you bring your own models through an OpenAI-compatible endpoint, and it speaks the Model Context Protocol (MCP). Both hooks let Tokonomix plug in: a single Tokonomix key gives Cursor access to 132 models across families, and the MCP server adds a cross-family consensus tool the Agent can call on high-stakes changes.

Cursor is one of the most widely used AI coding tools, and its Agent does real work: it plans a task, edits across many files, runs terminal commands, and iterates until tests pass. The more autonomy you hand it, the more a single confident-but-wrong model decision early in a run can cascade through everything that follows — which is exactly where a second opinion earns its keep.

Why route Cursor through Tokonomix?

Cursor ships excellent first-party models, but routing the chat and Agent layer through Tokonomix adds three things one built-in model can’t: breadth (every major family behind a single key), a verification step for the decisions that matter, and full cost and region control. You keep Cursor’s editor and its Tab autocomplete exactly as they are — Tokonomix changes only which model answers your chats and drives the Agent.

  • Cross-family consensus in the Agent. Add the Tokonomix MCP server and Cursor's Agent can call tokonomix_consensus_ask on a risky refactor — 2–6 frontier models cross-checked by a judge — instead of trusting one model.
  • One key, 132 models. Pick any model from the live catalogue in Cursor's chat without separate provider accounts.
  • Sovereignty routing — your choice, not EU-only. Pin models by where they run (hosting_region) or where they're made (origin_country) — EU-only, US-only, a single country, or a custom set. Enforce it per call, or set a predefined rule (e.g. auto-select EU models) in your dashboard. EU/GDPR is just one example.
  • Transparent per-call cost. Every call is metered in your Tokonomix dashboard.
  • Provenance & sovereignty. Filter by origin_country to choose models by where they were built — e.g. ?origin_country=FR or a CSV like FR,DE — when procurement rules require it.

Install Cursor — macOS, Windows, Linux

Cursor is a desktop app — download it from cursor.com (there is no CLI install). Open it, sign in, and open your project.

⊞ Windows
installer (.exe)
Download from cursor.com
→ run the Cursor installer
macOS
universal .dmg
Download from cursor.com
→ Intel + Apple Silicon
🐧 Linux
.AppImage
Download from cursor.com
→ chmod +x, then run

Connect Tokonomix

Cursor reaches Tokonomix two complementary ways. Route A puts Tokonomix models in Cursor’s chat; Route B lets the Agent call consensus as a tool. Use Route A to chat with a specific model, Route B when you want a verdict from several — and you can (and usually should) set up both at the same time; they serve different purposes and don’t conflict.

Cursor limitation — read first

Per Cursor's docs, custom API keys only work with chat models; Tab completion keeps using Cursor's built-in models and cannot be routed through Tokonomix. So Tokonomix powers Chat and the Agent (when it uses a chat model) — not inline Tab autocomplete.

Route A — Tokonomix model in Chat

Settings → Models → add a model, set your key and override the OpenAI Base URL. Selecting tokonomix-consensus as the model gives consensus answers in Chat directly — no MCP needed:

Base URL : https://tokonomix.ai/api/v1
API Key  : tok_live_your-key
Model    : tokonomix-consensus
# or any slug from GET /api/v1/models

Route B — consensus via MCP (Agent)

Add the Tokonomix MCP server — ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "tokonomix": {
      "command": "npx",
      "args": ["-y", "tokonomix-mcp"],
      "env": { "TOKONOMIX_API_KEY": "tok_live_..." }
    }
  }
}
Or just ask — no manual config

You can add the MCP server from Settings → MCP → Add new MCP server, or just tell Cursor’s Agent: "Add the Tokonomix MCP server — npx -y tokonomix-mcp — to my .cursor/mcp.json with my key." The Agent edits the file for you. After mcp.json is saved, reload or restart Cursor so the new tools appear in the Agent.

No key yet?

Create a free account at tokonomix.ai/dashboard/signup — it includes a free starter balance — then paste the key into Cursor (Route A) or your mcp.json (Route B).

The Tokonomix MCP server exposes tokonomix_consensus_ask (a multi-model verified answer), tokonomix_single_ask, tokonomix_list_models and tokonomix_get_skill. In practice you let Cursor’s Agent run a fast chat model for everyday edits and reach for the consensus tool only where being wrong is expensive — a schema migration, an auth flow, a public API contract.

What Tokonomix powers in Cursor

Cursor featureTokonomix supportHow
Chat (Cmd+L)YesCustom model + base URL (Route A)
Agent / ComposerYesChat model via base URL, plus tokonomix_consensus_ask via MCP (Route B)
Tab (autocomplete)NoStays on Cursor's built-in models — custom keys don't apply
MCP toolsYesconsensus, single-ask, model list, skills — in the Agent

Verify it works

In Chat, select tokonomix-consensus and ask a question — you should see a metered call in your Tokonomix dashboard. For the Agent, add the MCP server and ask it to "get a consensus verdict" on a change; the result comes back as the MCP tool result. Tab autocomplete keeps working on Cursor’s own models, unchanged.

In short

Route A gives Cursor’s chat 132 Tokonomix models on one key; Route B gives the Agent a consensus tool. Tab stays Cursor-native — by Cursor’s design, not a Tokonomix limit.

Model selection

Which model and endpoint should my agent use?

Pick by what your agent does, then swap the model id any time — every model in the catalog works through the same key.

What your agent doesUse this
Edits files / calls tools and streams responses (most coding agents)An Anthropic model via the Anthropic endpoint (tools + streaming both work)
Calls tools but does not need streaming (any model)The OpenAI-compatible endpoint with a single model id
Wants tools and streaming on a non-Anthropic modelRun non-streaming for now — streaming tool-calls on other providers is not enabled yet
Plain chat, no toolsEither endpoint; streaming is fine
Wants a cross-model verdict (consensus) on a high-stakes answerSet model to tokonomix-consensus on the OpenAI-compatible endpoint — note: consensus rejects tool requests (no tool-calling)
Endpoint & API reference (technical)

anthropic-messages → https://tokonomix.ai/api/anthropic — tools ✓, streaming ✓ (use an Anthropic model). openai-completions → https://tokonomix.ai/api/v1 — tools ✓ (single-model passthrough), streaming with tools not yet enabled. Plain chat: streaming works on both. Consensus: openai-completions, model tokonomix-consensus, no tools. Swap the model id to any slug returned by GET /api/v1/models.

Browse all models →

Related agents