Grok Build is an agentic coding CLI from xAI that runs in the terminal. After reading your codebase, it first presents a work plan describing what it will change and how, and carries out edits across multiple files only after you approve. The beta launched in May 2026, and the source code is published on GitHub under the Apache 2.0 license. Besides the interactive TUI (terminal interface), it supports headless execution from scripts and use from editors that support the Agent Client Protocol (ACP).
Key Features
- Plan-review workflow: Before rewriting any code, it shows a work plan and starts editing only after you approve it. Because you can check the scope and order of changes in advance, large refactors are less likely to drift off course
- Up to 8 parallel sub-agents: Each sub-agent gets its own Git worktree (an isolated copy of the working tree), and up to eight can run at once. Multiple tasks move forward in parallel while staying isolated so their changes do not collide
- MCP and ACP integration: You can connect MCP servers so the agent can use external tools. It reads existing MCP settings such as
~/.claude.json,.cursor/mcp.json, and a project’s.mcp.json, which makes it easy to reuse setups built for other tools. It can also be called from ACP-compatible editors - Reads existing instruction files: It loads project instruction files such as
AGENTS.mdand works according to the repository’s conventions - Headless execution: It can be run as a one-shot command, such as
grok -p, and embedded in CI/CD pipelines or internal scripts - Cross-session memory and extensions: A memory feature (
/memoryand others) records project decisions in the background and carries them over to the next session, and a marketplace distributes skills and plugins
Installation is a one-line command: curl on macOS / Linux and PowerShell on Windows.
Pricing
There is no standalone monthly plan for Grok Build. It is included in paid Grok subscriptions or used through pay-as-you-go xAI API billing. Because the official pricing page could not be fetched automatically and the amounts could not be confirmed from primary sources, the figures below are secondary information as of September 2026 (consistent across several comparison and explainer articles).
| Plan | Price | Notes |
|---|---|---|
| SuperGrok | $30/month ($300/year) | Reported as the entry-level plan that includes Grok Build |
| X Premium+ | $40/month | Reported to include Grok Build alongside X perks |
| SuperGrok Heavy | $300/month | The plan that first received the early beta. Reported to have the largest usage allowance |
| xAI API (pay-as-you-go) | $2.20 input / $6.60 output (per 1M tokens) | Grok 4.7 rates for up to 200K tokens. Usage beyond that is billed at higher rates |
Pricing is as of September 2026 (the table above is secondary information). Eligible plans and prices may change, so please check the official website for the latest pricing.
Pros & Cons
✅ Pros
- You can review the plan and stop it before execution, which helps prevent unintended mass rewrites and lightens the review burden
- Up to eight independent tasks can run at the same time, cutting total waiting time
- Because it reads existing MCP settings and
AGENTS.mdas-is, switching from another coding agent requires little setup - Being open source, you can verify its behavior in the code or modify it for your own environment
- Several comparison articles praise its response and execution speed, and it is said to suit work built on quick, repeated iterations
⚠️ Cons
- With no standalone plan, you need a paid Grok subscription or pay-as-you-go API billing
- The beta only began in May 2026, so know-how and surrounding resources are still thinner than for tools like Claude Code
- The service is still in flux, with the default model switching within a short period, so usability and pricing may change
- Some comparison articles report that it trails Claude Code and Codex CLI on certain benchmarks, so accuracy on hard tasks can vary by use case
Comparison with Similar Services
| Service | Difference from Grok Build |
|---|---|
| Claude Code | Also a terminal-based coding agent. It has a large user community and ecosystem of tools, and is regarded as strong at understanding large codebases |
| Codex CLI | OpenAI’s coding CLI. It emphasizes benchmark results and fast iteration |
| Cursor | Works entirely inside an editor (IDE). Grok Build, by contrast, is terminal-based and can be used regardless of editor |
Who Is It For
- People who already subscribe to SuperGrok or X Premium+ and want to try a coding agent without an extra contract
- People who want to check what the AI will do at the planning stage before handing over large changes
- People who want to push forward multi-module refactors or test additions at once with parallel agents
- Development teams that want to build AI-driven code fixes and reviews into CI or internal scripts
- People who want to compare another option while reusing MCP settings and instruction files made for other coding agents
Summary
Making “review the plan, then delegate” the standard flow is Grok Build’s defining trait. It also covers the features teams want, such as parallel sub-agents and headless execution. On the other hand, it is still early after launch, and the default model and eligible plans are prone to change. A good approach is to first test the plan quality and speed on a small repository of your own, compare it with the agents you normally use, and then decide whether to adopt it in earnest.