Skip to content
InferCPI logo

InferCPI MCP server

Which MCP servers are InferCPI's own, from the official registry and from asking its own domain, whether they answer, and what they offer. A program checked each of these, nobody copied them from a product page.

read 2026-10-29

← Back to InferCPI

Whose server it is
  • Registry namecom.infercpi/apiverified in the MCP registry

The registry name sits under infercpi.com. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.

When we asked itchecked 2026-10-29
  • Stateanswerschecked 2026-10-29
  • Tools10
  • Server calls itselfinfercpi
  • Server version1.23.3
  • Protocol version2025-06-18

On 2026-10-29 the server answered initialize and tools/list without a sign-in. First answer in 521 ms.

Should an agent connect itNot proven either way yet
  • Whether its tools do the product's job has not been judged yet.
  • Checked on 4 of the 7 days it takes to judge reliability.

Once connected

  • Parameters are untyped or undescribed in 3 of 10 tools: check arguments before calling.

Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.

Measured dailydaily checks: 4, the first on 2026-10-29
  • INITE MCP scoreafter a weeka score needs 7 daily checks: 4 so far
  • Answeredmeasuring4 of the 7 daily checks it takes to print a share
  • Median first answer503 ms
  • Space in an agent's context≈ 1,008 tokensthe definitions of all its tools (10), which an agent carries on every turn
  • Tool definitionsB · 78 of 100
  • Typed results0 of 10 toolsdeclared outputSchema: results come typed, read once
  • Parameters without a description: in 3 of 10
  • Not saying whether it only reads or can destroy: 10 of 10

What an agent can do with it

  • can do · Compare model list prices per million tokens over time
  • not sure · Find cheaper third-party hosts for open-weight models
  • not sure · Rank models by cost per correct answer

Each of the product's use cases, put to Jev against this server's tool list: can an agent do it with these tools?

Worth a look

None of the five warning signs in these definitions.

A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (25%), doing the product's job (25%), definitions (20%), context (10%), use (10%) and how readily a client connects (10%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →

Tools it offers10 tools · read 2026-10-29
  • index_latest

    Latest value of every index series (list-price chained/composition/hedonic, usage-weighted paid price, Törnqvist, OpenRouter shares, TPI replica) with dates, the latest editorial note, and the licence/attribution line. Start here.

  • index_series

    Full weekly history of one series as [[unix_ts, value], ...] plus its method string. signal: price_index | weighted_index | or_rankings | implied_margin | self_host | task_cost | hf_downloads | gpu_spot | gpu_neocloud | aws_gpu | speed |…

  • index_catalogue

    Every series the index publishes: signal/key, point count, first and last date, method string.

  • basket

    Every model priced in this week's list-price index: provider, blended $/1M (0.75 in + 0.25 out), Epoch capability score where matched.

  • events

    Priced events in first-party price lists, newest first: launch, price_cut, price_increase, removed, each with key, provider, old/new blended $/1M and pct. since: ISO date; kind: filter to one kind.

  • speed

    Latest daily speed run, measured by the index: per model and route (openrouter default routing, openrouter_pinned to the first-party host, anthropic and openai direct), median time to first token (ms), visible output tokens per second…

  • model_price

    Blended list $/1M for one model by any common name ('Claude Opus 5.5', 'anthropic/claude-opus-5.5', 'gpt-5.5'); matches by the index's name normaliser and returns every basket entry that matches.

  • open_vs_closed

    Cheapest blended $/1M at each capability level (Epoch ECI 130/145/160) for closed models, open models at their lab's price and open models at the cheapest third-party host; the open discount; DeepSeek's floor lift (cheapest price without…

  • self_hosting

    What it costs per 1M tokens to leave the APIs: desk boxes (Mac Studio lease, RTX 5090 bought) at full and 25% duty, H100 rented by source (Vast, AWS, RunPod, CoreWeave, Lambda, Nebius) and owned, against the API price of the same open…

  • cost_per_task

    Monthly measurement: dollars and tokens per completed task for the most-used models (fixed 8-task suite via OpenRouter at default settings, charged cost), reasoning share of output tokens, accuracy, and medians.

How to connect

Add the server to your MCP client config. Claude Desktop, Cursor and most other clients read this shape:

{
  "mcpServers": {
    "api": {
      "type": "http",
      "url": "https://infercpi.com/mcp"
    }
  }
}

Add it

Claude Code, in a terminal

claude mcp add --transport http infercpi https://infercpi.com/mcp

Add to CursorAdd to VS Code

claude.ai, ChatGPT and other clients: add https://infercpi.com/mcp as a custom connector in the settings.

Each line is something a program found when it went looking: the registry entry, the server's own answer, a spec that parses. A missing line means we did not find it, not that it does not exist.