<MCP.Playground/>
online

Grok 4.7

Runs MCP tools

SpaceXAI’s newest flagship, released 21 September 2026 and aimed at coding and knowledge work. SpaceXAI says it works longer on hard tasks and checks its own work more carefully, which is what multi-step tool use needs.

Vendor
SpaceXAI
Released
21 September 2026
Context
500K tokens
Input
Text, image, file

Paste a server URL, pick a model, and watch it call your tools in a real conversation. No install.

No server of your own? Leave it blank and use one of the public mock servers.

In your browser

MCP Agent Studio

Server has a public URL? Chat with Grok 4.7 on it, watch every tool call, and compare models side by side.

Open MCP Agent Studio →

On your computer

MCP Playground Desktop

Server runs on localhost or over stdio? The browser can’t reach it. The desktop app can, with Grok 4.7 in Agent Studio. Download free

Mac (Apple Silicon)Mac (Intel)Windowsor Microsoft Store

MCP Evals

Can Grok 4.7 actually complete tasks with your server?

One chat shows the tool calls work. An eval writes real tasks from your tool schemas, has Grok 4.7 drive each one, and reports where it got the wrong answer even though every call succeeded. Pick Grok 4.7 as the driver.

Run evals →See a sample

What Grok 4.7 is

Grok 4.7 succeeds Grok 4.6 as SpaceXAI’s model for coding, agent tasks and knowledge work. It launched in Cursor and Grok Build and is on the Grok API, model routers and cloud platforms. It has a 500K-token window, the same as Grok 4.6, and takes text, image and file input. A faster serving mode is also offered on the Grok API.

What SpaceXAI says

  • CursorBench 4.0: 46.3%, against 40.4% for Grok 4.6, 41.7% for GPT-5.6 Sol and 51.8% for Fable 5.1.
  • DeepSWE v1.1: 71.0% at high effort, against 70.0% for Fable 5.1 and 72.7% for GPT-5.6 Sol.
  • Terminal-Bench 4.0: 37.6%, against 20.3% for Grok 4.6 and 57.9% for Fable 5.1.
  • EEBench: 64.0%, against 56.4% for Fable 5.1 and 39.4% for GPT-5.6 Sol.

These are the vendor’s own claims, not measurements of ours. Run the model against your own server to find out whether they hold for your tools.

How it reaches MCP

  • The Grok API’s Remote MCP Tools. You pass a server URL and SpaceXAI’s side connects to it, through its own SDK or the OpenAI-compatible Responses API.
  • Cursor, where it launched, which acts as an MCP client
  • MCP Playground’s Agent Studio: paste your server URL above and run Grok 4.7 against it in the browser, with no SpaceXAI key

What to watch for

  • Remote MCP Tools on the Grok API accept only Streamable HTTP and SSE servers, not stdio, and the server has to be reachable from SpaceXAI’s side. A local server works here or in a local client but not through that route.
  • The window is 500K tokens, half of Grok 4.3’s 1M. Long tool transcripts that fit on older Grok models can hit the limit on 4.7.
  • SpaceXAI’s own table shows a wide Terminal-Bench 4.0 gap to Fable 5.1 (37.6% against 57.9%) next to near-parity on DeepSWE. Long terminal-style chains are where to check it.

Run Grok 4.7 and Grok 4.6 on the same four-step task against the complex-schema mock: create_user_profile, process_order, analyze_data, then configure_workflow. SpaceXAI says 4.7 checks its own work more carefully. A chain of dependent calls is where that should show. All the mock servers →

Available here

ModelModel ID
Grok 4.7x-ai/grok-4.7

Go deeper

Frequently asked questions

When was Grok 4.7 released?
21 September 2026. It launched in Cursor and Grok Build and on the Grok API the same day. You can run it in MCP Playground against your own server.
Does Grok 4.7 support MCP?
Yes. The Grok API can connect to remote MCP servers itself over Streamable HTTP or SSE, and any MCP client can drive it through function calling. You can run it against your own server here without a SpaceXAI key.
Is Grok 4.7 better than Claude Fable 5.1?
On SpaceXAI’s own table it leads Fable 5.1 on EEBench and the Harvey legal agent benchmark and trails on CursorBench 4.0 and Terminal-Bench 4.0. Which one handles your tools better is something only a run against your server will show.
What changed from Grok 4.6?
SpaceXAI reports gains across its coding and agent benchmarks, including CursorBench 4.0 from 40.4% to 46.3% and Terminal-Bench 4.0 from 20.3% to 37.6%, and a new safeguard stack. The window stays at 500K tokens.

Sources