Read-only community data · No account

Ask ChatGPT about Tokrate’s community measurements.

Connect a custom MCP app to browse published coding-model response-speed comparisons, cohorts and history in a conversation.

The service reads Tokrate’s public aggregates. It does not measure the speed of your current ChatGPT or local coding session. You do not need to install the desktop companion.

MCP endpoint

https://tokrate.dev/api/public/v1/mcp

No authentication is required. The same endpoint is compatible with Codex and other MCP clients.

Download Tokrate plugin ZIP

What you can ask

Compare response speed

Compare published model and provider results for a selected time window, with sample counts and clearly identified model/provider groups.

Explore history and signals

Review available history and coverage signals to understand how much published evidence supports a comparison.

Check the methodology

See how measures are defined. Efficiency is reported separately from response speed; neither measures answer quality.

These are community observations from real coding sessions, not a controlled benchmark or a report about your own local session.

Prompts to try

Compare providers

Compare published response speed for two coding models with observations over the last 7 days. Show an interactive panel, keep each provider separate, and explain the sample counts and comparison limits.

Find published results

Which coding models have published observations for the last 7 days? Show model and provider, sample counts and freshness. For one model, inspect the separate coding-tool and reasoning-effort cohorts.

Understand the measures

Explain how Tokrate measures response speed, turn speed and first-token wait. What does the separate token-efficiency indicator mean, and what can these community observations tell me?

Connect from ChatGPT

In ChatGPT on the web, open Plugins → Add → Create custom MCP server. In the earlier Apps interface, enable Developer mode in Settings → Apps → Advanced settings if available.
Name the connection Tokrate and paste the endpoint above. Select No authentication, scan the tools, then create the app.
Start a chat, choose Tokrate from the Plugins or Apps menu, and ask one of the prompts above.

Access and menu names vary by plan and workspace; a workspace administrator may need to enable developer mode. See OpenAI’s ChatGPT developer mode guide for current details.

This page describes direct setup as a custom MCP app; it does not imply a public directory listing or approval.

Privacy

Tokrate tools read published aggregate results only. Requests may include a selected model, provider or time window; those arguments are processed transiently to answer the query. Existing service rate limits and operational logging apply. Tokrate does not store full chats, files or personal telemetry through these tools. ChatGPT handles conversations under its own policies. Read the Tokrate privacy notice.

Support

Questions about the Tokrate MCP service? Contact [email protected].