How much does Grok Build cost?

Available on Grok plans, from a free tier to SuperGrok at $30 a month, with Business and Enterprise through sales. On an xAI API key it bills per token: Grok 4.7 at $2 per million input tokens, $0.50 cached and $6 output for prompts under 200,000 tokens, and $4, $1 and $12 above that; Grok Build 0.1 at $1, $0.20 and $2, doubling above 200,000 tokens. The US regional endpoint costs 10 percent more.

What are the best alternatives to Grok Build?

Based on documented capability coverage across the Agentic Index 14 point taxonomy, the closest alternatives to Grok Build among coding agents are OpenCode, Amp, Augment Code, Bubble. Each is graded on the same public evidence, and no vendor pays for placement.

Does Grok Build support MCP and tool calling?

Full under the Agentic Index verification standard. Grok Build publishes public evidence of Model Context Protocol (MCP) support, tool calling, and API or SDK extensibility for connecting agents to external tools. Evidence basis: Other software drives Grok Build through the Agent Client Protocol, which runs it as a JSON RPC server over standard input and output, and through headless mode, which takes a prompt and returns plain, JSON or streaming JSON output for scri. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does Grok Build have SOC 2, SSO and role based access control?

Full under the Agentic Index verification standard. Grok Build publishes public evidence of enterprise security and identity controls, covering a compliance posture such as SOC 2 or ISO 27001, single sign on (SSO or SAML), and role based access control (RBAC). Evidence basis: Enterprises sign developers in through OIDC with Entra ID, Okta or Auth0, pin logins to the company team, turn off API key sign in and lock settings through a device management file that overrides every other layer, with allowed models and. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Can Grok Build be self hosted or run in a private VPC?

Full under the Agentic Index verification standard. Grok Build publishes public evidence of self hosted or private VPC deployment, along with data residency options for controlling where agent data is processed and stored. Evidence basis: The agent and its tools run on the developer's own machine, sending inference through xAI's proxy unless pointed elsewhere, and the open source build can run fully local first against the customer's own inference endpoint. Session sync to x. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Back to vendors

Grok Build

Q: Who is Grok Build for?

developers who work in the terminal, engineering teams running coding agents in CI, enterprises standardizing a coding agent under central configuration, teams moving from Claude Code or Cursor conventions

Also known as: xAI Grok Build, Grok Build CLI, x.ai/cli

Visit site

Coding agentfirst party productVerified 2026-11-08

Entry priceComes with Grok plans, from a free tier to SuperGrok at $30 a month, or runs on an xAI API key at per token rates: Grok 4.7 at $2 per million input tokens and $6 per million output.Full pricing detail

Grok Build is xAI's coding agent for the terminal, for scripts and CI, and for other editors over the Agent Client Protocol. It plans, edits and runs code with subagents, background tasks and git worktrees, behind permission modes, plan approval and sandbox profiles, and extends through MCP, hooks, skills and plugins compatible with Claude Code. Grok 4.7 is the default model, any compatible endpoint can stand in, and it comes with Grok plans or runs on an API key.

Grok Build is the coding agent from xAI, the maker of the Grok models, which SpaceX acquired in February 2026; xAI's site footer now carries the name SpaceX Super Intelligence LLC. It launched in early beta in May 2026, was released as open source in July, and since August has been available on every Grok plan, on the web and on mobile as well as in the terminal, though xAI still labels it beta.

It runs as an interactive terminal app, headless in scripts and CI with plain, JSON or streaming JSON output, or as an Agent Client Protocol server inside other editors. The agent reads and edits code, runs commands, searches the web, and splits work across subagents for general work, exploration and planning, with background tasks, recurring loops, long running goals and git worktrees for parallel changes.

Plan mode keeps it to a plan file until a person approves, permission modes run from asking before each action to automatic approval through a classifier, and sandbox profiles on Linux and macOS limit what it can touch. MCP servers, hooks on 14 events, skills and a plugin marketplace extend it, and it reads Claude Code and Cursor configuration, rules and hooks so teams can switch without rewriting them. Memory across sessions is an experimental option that is off by default.

Grok 4.7 is the default model, and users can switch models or point the agent at any OpenAI, Anthropic or Responses compatible endpoint, including their own inference. On an xAI API key, Grok 4.7 costs $2 per million input tokens and $6 per million output tokens, and xAI's own Grok Build 0.1 model $1 and $2; on subscriptions it comes with Grok plans, from a free tier to SuperGrok at $30 a month. Enterprises get OIDC sign in through Entra ID, Okta or Auth0, team pinned login, configuration locked through device management, and zero data retention by team, under xAI's SOC 2 Type 2 report.

It fits developers and teams who want a fast terminal agent they can script, lock down centrally and run on Grok or another model, including teams already working to Claude Code or Cursor conventions. It fits less well for buyers who want a codebase index, an evaluation harness, a browser tool or a central audit trail of agent sessions, none of which ship with it today, or who need a product that has left beta.

Vendor details

Canonical URL

https://x.ai/cli

Subcategory

Terminal coding agent from xAI that runs interactively, headless in CI or inside other editors over the Agent Client Protocol

Funding status

Product of xAI, which SpaceX acquired in February 2026; the site footer names SpaceX Super Intelligence LLC, and enterprise terms are with X.AI LLC of Palo Alto.

Company status

first party product

Use cases & customers

Primary use cases

agentic coding in the terminalheadless coding runs in scripts and CIrunning a coding agent inside other editors over ACPparallel changes with subagents and git worktreescentrally managed coding agent rollout with OIDC and locked settings

Target customers

developers who work in the terminalengineering teams running coding agents in CIenterprises standardizing a coding agent under central configurationteams moving from Claude Code or Cursor conventions

Deployment options

Local CLI on the developer's machineHeadless in CISelf hosted inference through a custom endpoint

Integrations

MCP servers over stdio or HTTP with OAuth, user or project scoped, loading existing Claude Code and Cursor MCP configurations. Hooks on 14 events as commands or HTTP calls, and plugins bundling skills, agents, hooks, MCP and language server tools, compatible with Claude Code plugins. Runs inside other editors over the Agent Client Protocol and in CI through headless mode. Enterprise sign in through Entra ID, Okta or Auth0.

In practice

A developer asks Grok Build to migrate a service from one logging library to another. It plans the change in plan mode, waits for approval, then sends an explorer subagent through the codebase while it edits files in a separate worktree.

A platform team runs Grok Build headless in CI on every nightly build to update dependencies, signing in with an API key and using a mode that accepts edits without prompts, then reviews the resulting branch in the morning.

A security team pins Grok Build to company sign in through Okta, turns off always approve, limits it to two approved models and enables zero data retention for the team, all through one managed configuration file pushed by device management.

Sources & related URLs

Related / legacy domains

https://docs.x.ai/build/overview.md https://github.com/xai-org/grok-build https://x.ai/news

Research sources

Agentic Index coverage score

10.5 / 14 capabilities · 75%

Integrations & Tool Calling	Full
MCP servers connect over stdio or HTTP with OAuth sign in, scoped to a user or a project, and existing Claude Code and Cursor MCP configurations load as they are. Hooks fire on 14 events as shell commands or HTTP calls, and plugins bundle skills, agents, hooks, MCP servers and language server tools. With those, the agent reaches any service a team exposes, alongside its own shell, file editing and web search tools. Sourcedocs.x.ai/build/features/mcp-servers.mdread 2026-11-08
Workflow Orchestration	Full
Work splits across subagents, a general purpose agent, an explorer and a planner, plus custom agents and personas a team defines. Background commands, subagents and monitors run while the main session continues, git worktrees keep parallel changes apart, /goal drives long runs toward an objective, and workflow files script multistep jobs. Headless mode lets CI chain Grok Build into larger pipelines of its own. Sourcedocs.x.ai/build/features/subagents.mdread 2026-11-08
Knowledge Grounding & RAG	Partial
Context comes from the working tree at run time: file search that respects .gitignore, an optional language server tool that is off by default, built in web search, and web fetch limited to an allowlist of domains. A deep research command runs a research workflow in the background. No index of the codebase or of company documents persists between sessions, so each run gathers what it needs fresh. Sourcedocs.x.ai/build/settings/reference.mdread 2026-11-08
Human Oversight & Guardrails	Full
The default permission mode stops for approval before each action, with an automatic mode that uses a classifier and an always approve mode that admins can lock off, and deny rules win over everything else. Plan mode limits the agent to its plan file until a person approves, even under automatic modes. Sandbox profiles for workspace, devbox, read only and strict use Landlock on Linux and Seatbelt on macOS, though the sandbox is off by default. Sourcedocs.x.ai/build/features/permissions.mdread 2026-11-08
Security, Identity & Governance	Full
Enterprises sign developers in through OIDC with Entra ID, Okta or Auth0, pin logins to the company team, turn off API key sign in and lock settings through a device management file that overrides every other layer, with allowed models and approval modes set centrally. Teams can switch on zero data retention, under which no trace or code is kept. xAI holds a SOC 2 Type 2 report, available through its trust center under NDA. Sourcedocs.x.ai/build/enterprise.mdread 2026-11-08
Observability & Auditability	Partial
Each developer's sessions are saved on their machine, searchable and exportable, and can be shared by link, with /usage and an agent dashboard showing activity, plus local log files for debugging. Telemetry can be switched off by policy. The xAI console's audit log and usage explorer cover API usage in general, and there is no team view or audit trail built around Grok Build sessions. Sourcedocs.x.ai/build/features/dashboard.mdread 2026-11-08
Memory & State Persistence	Partial
Project rules load from AGENTS.md, CLAUDE.md, .grok/rules and Cursor rules at global and nested scopes, and those are standing instructions. Sessions are saved locally and can be resumed, forked, rewound and exported. A memory feature that keeps markdown notes at workspace and global scope, leaves secrets out and can be cleared by scope sits behind an experimental setting that is off by default. Sourcex.ai/news/grok-build-memoryread 2026-11-08
Deployment & Data Residency	Full
The agent and its tools run on the developer's own machine, sending inference through xAI's proxy unless pointed elsewhere, and the open source build can run fully local first against the customer's own inference endpoint. Session sync to xAI's cloud is optional. On the xAI API, a US endpoint guarantees US processing for Grok 4.7 and 4.6 at a 10 percent premium, while the global endpoint carries no region guarantee. Sourcedocs.x.ai/developers/advanced-api-usage/regions.mdread 2026-11-08
Prebuilt Agents / Templates / Packs	Full
A built in marketplace offers skills and plugins, compatible with those written for Claude Code, and /skillify turns a finished session into a reusable skill. Three subagents ship ready to use, for general work, exploring a codebase and planning, and teams can add custom agents and personas alongside them. Each plugin or skill installs on its own, so a team adopts only the pieces it needs. Sourcedocs.x.ai/build/features/skills-plugins-marketplaces.mdread 2026-11-08
Triggers & Channel Coverage	Full
Beyond a person typing, work arrives from scripts and CI through headless mode, which signs in with an API key and runs in modes that need no prompts. The /loop command schedules a recurring task every 60 seconds or more, up to 50 at once, each expiring after seven days, and background monitors watch for conditions while a session runs. There are no inbound webhooks, so events outside the terminal reach it through the customer's own CI. Sourcedocs.x.ai/build/features/background-tasks.mdread 2026-11-08
Model Flexibility & Routing	Full
Grok 4.7 is the default model, and users switch with /model or a command line flag, with Composer 2.5 also on offer. Any other model works through a base URL and a chat completions, responses or messages backend, which covers OpenAI and Anthropic compatible providers and self hosted inference. Admins limit the choice with an allowed models list in the managed configuration. Sourcedocs.x.ai/build/settings.mdread 2026-11-08
APIs / SDKs / MCP Extensibility	Full
Other software drives Grok Build through the Agent Client Protocol, which runs it as a JSON RPC server over standard input and output, and through headless mode, which takes a prompt and returns plain, JSON or streaming JSON output for scripts. The code is open source on GitHub. There is no separate SDK, and model calls go through the xAI API or any endpoint the customer configures. Sourcedocs.x.ai/build/cli/headless-scripting.mdread 2026-11-08
Testing, Debugging & Optimization	Not documented
Grok Build has no evaluation harness, saved test suite or scored comparison of agent runs. It can run a project's own tests from the shell like any other command, which uses the customer's test suite as the check. xAI's published results, such as Grok 4.7's scores on public coding benchmarks, measure the model rather than this agent, and nothing in the product scores a change before it ships. Sourcedocs.x.ai/build/overview.mdread 2026-11-08
Browser / Computer-use	Not documented
No browser or computer use tool ships with the agent. It works through the shell, file edits, web search and an allowlisted web fetch, which read pages as text and never click through them. Browser automation appears only as an example of an MCP server a team could add itself, so any browser control would come from a third party server that the team installs, configures and runs on its own machines. Sourcex.ai/cliread 2026-11-08

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Comes with Grok plans, from a free tier to SuperGrok at $30 a month, or runs on an xAI API key at per token rates: Grok 4.7 at $2 per million input tokens and $6 per million output.

Flat monthly Grok plans with usage limits, or per million input, cached and output tokens on an xAI API key, with higher rates above 200,000 token prompts.

Free tier

Included quota

Usage limits follow the Grok plan; the free tier includes Grok Build but not Grok 4.7 Fast.

What is public

Per token API rates for Grok 4.7, Grok 4.7 Fast and Grok Build 0.1, the long context surcharge, the US endpoint premium, and the Free and SuperGrok plan prices.

Billing mechanics

Either a Grok subscription billed monthly, or prepaid or invoiced xAI API usage on a key that Grok Build signs in with. Business and Enterprise plans are contracted through sales.

Cost watchouts

Long context prompts above 200,000 tokens double the per token price. The US regional endpoint adds 10 percent. Grok 4.7 Fast costs twice Grok 4.7 per token.

Variable cost rationale

On a subscription the price is flat with usage limits set by plan. On an API key every token is billed, at double the rate once a prompt passes 200,000 tokens, so long sessions in large repositories cost more per turn. Choosing Grok Build 0.1 halves the input price against Grok 4.7.

Additional watchouts

Which subscription plans include Grok Build differs between the product page and a later announcement, and the plan grid's Grok Build row could not be read.

Overage / add-ons

On an API key there is no plan ceiling; prompts of 200,000 tokens or more bill at double the base rate.

Sales call required

Mixed (some tiers require a call)

Free / trial

A free tier includes Grok Build, with usage limits set by plan.

Lowest paid plan

SuperGrok at $30 a month, or pay per token on an xAI API key

Commercial notes

Grok Build comes from xAI, which SpaceX acquired in February 2026; the site footer names SpaceX Super Intelligence LLC. It is open source and still labeled beta.

Key ambiguities

The product page says Grok Build is for SuperGrok and X Premium+ subscribers, while an August announcement says it is on every plan and the docs describe a free tier.

Cancellation / refund

Monthly subscriptions; API usage billed per token; Enterprise by contract.

Support SLA / resale

Enterprise adds OIDC sign in, team pinned login, managed configuration and zero data retention by team.

Missing data

Prices for SuperGrok Heavy, Business and Enterprise, and the usage limits each plan sets for Grok Build.

Agentic Index verified 2026-11-08

Official pricing page Docs

Related vendors

Alternatives to Grok Build

The closest documented capability profiles to Grok Build among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.

OpenCode10.5 / 14Adds documented Testing, Debugging & Optimization
Amp12.0 / 14Adds documented Testing, Debugging & Optimization and Browser & Computer Use
Augment Code12.0 / 14Adds documented Testing, Debugging & Optimization
Bubble10.0 / 14Adds documented Testing, Debugging & Optimization
DeepSource12.0 / 14Adds documented Testing, Debugging & Optimization
Poolside11.0 / 14Adds documented Testing, Debugging & Optimization

Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded

Compare Grok Build with other vendors →Compare all coding agents →See where Grok Build ranks →AI agents for software development →

Data confidence: medium