CheapestInference MCP Server

Remote MCP server for CheapestInference — unlimited LLM inference at a flat monthly price. No token caps, no per-token billing: subscribe to a pool of frontier open-source models and use them through an OpenAI-compatible API (an Anthropic-compatible endpoint is included, so it works with Claude Code, Cursor, and any OpenAI SDK).

This MCP server is the commercial front door. From any MCP client — or from an autonomous agent — you can browse live pools and pricing, get a USDC quote, pay on Base L2, and receive an API key, end to end, without a browser and without a human in the loop.

https://api.cheapestinference.com/mcp

Transport: Streamable HTTP, stateless. No authentication to connect — management tools take your API key as a tool argument.

Note: inference itself is served over the OpenAI-compatible base URL (https://api.cheapestinference.com/v1), not over MCP. The get_connection_info tool hands you ready-to-paste client snippets.

Quick start

Claude Code

claude mcp add --transport http cheapestinference https://api.cheapestinference.com/mcp

Cursor / generic JSON config

{
  "mcpServers": {
    "cheapestinference": { "url": "https://api.cheapestinference.com/mcp" }
  }
}

Tools

Tool	Auth	What it does
list_pools	none	Active pools with per-block USDC prices (live) and availability
get_pool	none	One pool in detail: models, daily blocks, what the subscription includes
get_crypto_quote	none	Exact USDC amount, payment address on Base, and the claim message to sign
confirm_crypto_subscription	wallet signature	Verifies your payment on-chain and returns your API key
get_card_checkout_link	none	Dashboard URL to finish a card subscription in a browser
get_subscription_status	API key	Pool, reserved blocks, expiry, renewal semantics for your key
get_connection_info	optional API key	Base URLs + client snippets (OpenAI SDK, Claude Code, curl)

Subscribing with USDC, end to end

1. list_pools → pick a pool and one or more daily 8-hour blocks (all three = 24/7).
2. get_crypto_quote → exact USDC amount, the payTo address on Base, and a claim-message template.
3. Send the USDC (ERC-20) transfer from your wallet.
4. Sign the claim message (EIP-191, transaction hash filled in) with the same wallet that paid.
5. confirm_crypto_subscription with tx_hash, pool_slug, blocks, signature → your API key for the Unlimited pool.

USDC subscriptions are one-time 30-day purchases — renew by repeating the flow. Card subscriptions (via get_card_checkout_link) auto-renew until cancelled.

Agents with an x402-capable wallet (Coinbase AgentKit, x402-fetch) can also skip subscriptions entirely: keyless requests to the inference endpoint return a 402 with USDC pricing for pay-per-request.

Links

• Website: cheapestinference.com
• Docs: docs.cheapestinference.com/guides/mcp
• Agent subscriptions (REST): docs.cheapestinference.com/guides/agent-subscriptions
• x402 pay-per-request: docs.cheapestinference.com/guides/x402
• Official MCP Registry: com.cheapestinference/mcp
• Directories: Smithery · LobeHub · Glama · mcp.so