MCP server

The same prices, as tools an AI agent can call. Point Claude, Cursor or any MCP client at https://mcp.indextkn.com/mcp and it can look up a price instead of guessing at one.

What it is

MCP (Model Context Protocol) is how an AI assistant talks to an outside service. The indextkn MCP server wraps this API in seven read-only tools, so an agent answering a question about cost reads today's published prices rather than whatever it remembers from training.

It is the same data, the same key and the same monthly limit as the REST API. Use the REST API when you are writing code. Use the MCP server when an agent is doing the asking. If you would rather teach an agent to call the REST API itself, with nothing running, install the agent skill instead.

Install

You need an API key. Create one on the dashboard, then pick your client.

Claude Code

Terminal

claude mcp add --transport http indextkn https://mcp.indextkn.com/mcp \
  --header "Authorization: Bearer itk_live_YOUR_KEY"

Claude Desktop, Cursor, and other clients

Add this to the client's MCP configuration file:

JSON

{
  "mcpServers": {
    "indextkn": {
      "type": "http",
      "url": "https://mcp.indextkn.com/mcp",
      "headers": { "Authorization": "Bearer itk_live_YOUR_KEY" }
    }
  }
}

Your key is a credential. It goes in the Authorization header, never in the URL. The server stores nothing: it forwards your key to this API for one request and discards it.

Tools

All seven are read-only and are marked as such, so most clients can run them without asking for permission each time.

Tool	What it answers
`list_models`	Which models exist and who sells each. Supports search, lab, provider, modality and paging. This is where an agent finds the exact ids the other tools take.
`get_model`	One model in full, with every provider offer: prices, context window, max output, status, source URL.
`get_prices`	The current price of every matching offer. Needs at least one of model, provider or lab.
`compare_providers`	The price spread for one model across providers: cheapest, dearest, median and the ratio between them.
`calculate_cost`	What a token workload costs on every provider that sells the model, cheapest first. Step-up tiers are applied automatically.
`list_providers`	The seller catalogue, with live counts and the kind of seller each is.
`list_labs`	The lab catalogue, meaning who built the models.

What to ask it

Questions that work well, in plain language:

Pick a provider. "Who is cheapest for gpt-oss-120b, and how much cheaper than the rest?" The spread is often 4x or more for the same model.
Budget a feature before building it. "We send 4k tokens in and get 800 out, 200k times a month. What does that cost on Opus 4.5 versus Sonnet, on every provider that sells them?"
Check what caching saves. "Three quarters of our prompt is a fixed system block. What does prompt caching change?"
Size a request honestly. "Which providers of this model actually serve a 200k context window?" The cheapest one often serves a smaller window than the lab does.
Build a cost-aware fallback chain. "Order these five models by output price and give me a router that fails over down the list."

Prices carry an accuracy status, and the tools pass it through. Ask the agent to say when a number is suspicious or partial rather than quietly trusting it, and to cite the source_url before you spend against a figure.

Running it locally

The server is open source and can run as a local process over stdio instead, with the key in the environment. The repository has the configuration and the security notes: github.com/indextkn/indextkn-mcp.