Open source · MIT license · MCP server

Find free LLM models.
Use them directly.

One MCP tool gives your agent instant awareness of what's available for free. No proxy, no middleware. Token Scout tells you where to go — you make the call.

View on GitHub

One tool. Every free model.

Find models by capability

token_scout(query="reasoning")
→ DeepSeek R1 671B-MoE on OpenRouter, Grok 3 Mini on GitHub...

Find the fastest option

token_scout(query="fast classification")
→ Llama 3.1 8B on Groq, Llama 4 Scout on Cerebras...

Rank by what matters

token_scout(query="", prefer="context")
→ all models ranked by context window size
prefer options: quota (most requests remaining), speed (fastest), context (largest window)

Check your status

token_scout(query="")
→ status: which providers are configured, how many models available

19 models. 6 providers. All free.

No credit card required. Sign up for the ones you want — Token Scout auto-detects your keys.

Provider	Models
Groq	Llama 4 Scout/Maverick, Llama 3.3 70B, Llama 3.1 8B, Kimi K2, Qwen3 32B, GPT-OSS 120B
Cerebras	Llama 3.3 70B, Llama 4 Scout, Qwen3 32B
Mistral	Mistral Small 3.1 24B
OpenRouter	Gemma 3 27B, Llama 3.3 70B, Mistral Small 3.1, DeepSeek R1
GitHub Models	GPT-4o, DeepSeek R1, Grok 3 Mini
Google AI	Gemini 2.0 Flash (1M context)

Five minutes to first query

Install

$ git clone https://github.com/jackccrawford/token-scout.git
$ cd token-scout
$ pip install -e .

Add your keys (whichever providers you have)

$ export GROQ_API_KEY=gsk_...
$ export OPENROUTER_API_KEY=sk-or-...
$ export GITHUB_TOKEN=ghp_...

Add to Claude Code

$ claude mcp add token-scout -- token-scout

Or add to Claude Desktop

// claude_desktop_config.json

"mcpServers":

"token-scout":

"command": "token-scout"

A registry, not a proxy

Token Scout knows what free models exist, what keys you have, and what each model is good at. When you query, it searches, filters, ranks, and returns everything you need to make the call yourself.

Your agent → Token Scout (registry) → model endpoint
Zero inference overhead. The call goes straight from you to the provider.

Returns: model ID, endpoint, API style, key env var, context window, strengths, and quota. Everything your agent needs to call the model directly. No middleman.

Why not just bookmark the docs?

You could. But your AI can't read your bookmarks. Token Scout gives any MCP-connected AI instant awareness of what's available for free. Instead of hardcoding model IDs, an agent can ask "what's good for reasoning?" and get a current answer.

It's also a single place to update when providers change their free tiers — update the registry, every connected AI benefits.

Your agent doesn't know
what's free.

One tool. Five minutes. 19 models.