superpower-codex-mcp
This server coordinates AI-driven software development workflows, integrating Codex for planning, review, debugging, and verification with Antigravity CLI for code execution.
validate_workspace: Checks workspace authorization, Git repository status, and Antigravity CLI availability without modifying files or invoking a model.summarize_repo_context: Reads repository instruction files (e.g.,AGENTS.md,GEMINI.md) to gather project context before coding.plan_with_codex: Asks Codex to generate and save a structured implementation plan before any code is written, with configurable reasoning depth.review_with_codex: Asks Codex to review a diff, working tree, or specific files, returning severity-ordered findings in structured JSON.debug_with_codex: Asks Codex to systematically analyze failures, stack traces, or command output without applying fixes.verify_with_codex: Runs explicitly allowed verification commands and asks Codex to assess the evidence before declaring completion.review_code_quality: Runs a local static analysis scan on TypeScript files (checking foras anycasts, empty catch blocks, hardcoded secrets,console.log, TODOs, etc.) without consuming any LLM tokens.run_antigravity_coding_task: Executes coding tasks via Antigravity CLI with strict execute/plan contracts, Git-verified commits, defined file scope, and acceptance criteria traceability.run_gemini_coding_task: Deprecated alias forrun_antigravity_coding_task, providing the same functionality with a deprecation notice.run_development_workflow: Orchestrates the full plan → implement → review → fix → verify pipeline automatically, with configurable iteration limits, skip flags, and verification commands.
Integrates with OpenAI's Codex CLI for planning, review, debugging, and verification tasks in a software development workflow.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@superpower-codex-mcpPlan and code the user registration feature"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
superpower-codex-mcp
A TypeScript MCP server that coordinates Codex and Antigravity CLI with a Superpowers-style software development workflow.
Codex handles planning, review, debugging, and verification. Antigravity CLI is the canonical coding executor. The server adds workspace authorization, Git-based change verification, acceptance criteria, compact task reports, and persisted diagnostics.
Requirements
Node.js 20.19 or newer, or Node.js 22.12 or newer
Codex CLI installed and authenticated
Antigravity CLI (
agy) installed and authenticatedGit
Related MCP server: Codex MCP Server
Install
git clone https://github.com/Pushplayhero/superpower-codex-mcp.git
cd superpower-codex-mcp
npm.cmd install
npm.cmd run build
npm.cmd testOn macOS or Linux, use npm instead of npm.cmd.
Codex Desktop configuration
Add the server to your Codex MCP configuration. Replace the paths with your local clone and the workspaces the server may access.
[mcp_servers.superpower-codex]
command = "node"
args = ["C:\\path\\to\\superpower-codex-mcp\\dist\\src\\index.js"]
[mcp_servers.superpower-codex.env]
SUPERPOWER_CODEX_ALLOWED_ROOTS = "C:\\path\\to\\projects"Restart Codex Desktop after changing MCP configuration.
On Windows, the server resolves Codex in this order:
SUPERPOWER_CODEX_COMMANDcodex.exefound onPATHAn existing global npm
@openai/codexentrypoint
The packaged Codex Desktop binary under WindowsApps\OpenAI.Codex_* is skipped
because Windows can reject external child-process launches with EPERM.
If no candidate exists, Codex-backed tools return structured diagnostics with
codexCliAvailable: false and the paths that were searched.
Antigravity configuration
In a project that should use this MCP server, add .agents/mcp_config.json:
{
"mcpServers": {
"superpower-codex": {
"command": "node",
"args": [
"C:\\path\\to\\superpower-codex-mcp\\dist\\src\\index.js"
],
"env": {
"SUPERPOWER_CODEX_ALLOWED_ROOTS": "C:\\path\\to\\projects"
}
}
}
}The default Antigravity executable is:
%LOCALAPPDATA%\agy\bin\agy.exeSet SUPERPOWER_ANTIGRAVITY_COMMAND when it is installed elsewhere:
$env:SUPERPOWER_ANTIGRAVITY_COMMAND = "C:\custom\path\agy.exe"Available tools
Tool | Purpose |
| Validate workspace authorization and local capabilities without invoking a model. |
| Read repository instructions such as |
| Ask Codex to create and save an implementation plan. |
| Ask Codex to review a diff or selected files. |
| Ask Codex to investigate failures systematically. |
| Run explicitly allowed verification commands and ask Codex to assess the evidence. |
| Run a local TypeScript structural scan without consuming model tokens. |
| Canonical coding execution tool backed by Antigravity CLI. |
| Deprecated compatibility alias for |
| Coordinate plan, implement, review, fix, and verify stages. |
Recommended workflow
Call
validate_workspace.Call
summarize_repo_context.Use
plan_with_codexfor broad or risky changes.Execute the approved task with
run_antigravity_coding_task.Use
review_code_qualityas a token-free pre-filter.Call
review_with_codex.Send review corrections back to Antigravity when required.
Call
verify_with_codexbefore declaring completion.
For an automated version of this sequence, use run_development_workflow.
Its verificationCommands input defaults to ["npm test", "npm run typecheck"]
and can be replaced for pnpm, Bun, Python, Go, monorepo, or other project
toolchains.
review_with_codex returns validated structured JSON with a status,
summary, and severity-ordered findings array. Malformed model output is
reported as an error instead of being interpreted as a clean review.
review_code_quality currently supports TypeScript. A Python-only workspace
returns unsupportedLanguage: "python" instead of presenting zero scanned
files as a clean review. Recursive Python detection ignores virtual environment
and build directories (such as venv, .venv, __pycache__, .tox, build,
dist, node_modules, and .git).
For explicit mixed selections (e.g. app.ts and app.py), the tool scans the
TypeScript files and returns unsupportedFiles: ["app.py"] while omitting
unsupportedLanguage.
Coding task example
{
"workspacePath": "C:\\path\\to\\project",
"prompt": "Implement the approved task using test-driven development.",
"allowExecution": true,
"timeoutSeconds": 1800,
"model": "Gemini 3.5 Flash (Medium)",
"mode": "execute",
"planApproved": true,
"requireCommit": true,
"requireCleanWorkspace": true,
"acceptanceCriteria": [
{
"id": "AC-1",
"description": "The new behavior is covered by a regression test."
}
],
"allowedFiles": [
"src/example.ts",
"tests/example.test.ts"
],
"responseDetail": "summary"
}Supported model names:
gemini-3.5-flashGemini 3.5 Flash (Medium)Gemini 3.1 Pro (High)
Unsupported legacy Gemini model names are rejected. The server does not automatically retry with a stronger model.
Status model and diagnostics
Coding task statuses
The coding tools (run_antigravity_coding_task and the legacy alias
run_gemini_coding_task) return a JSON payload with a status field:
planned: A requested planning task was completed.implemented_unverified: Files may be implemented, but required test or acceptance evidence is incomplete.tests_passed: All supplied acceptance criteria map to named passing tests, but no commit was required or verified.committed: Required commit and acceptance evidence were verified successfully.mode_mismatch: Execute mode was requested, but the model asked for approval or returned a plan-only response.contract_failed: Preconditions or postconditions failed (e.g., dirty workspace before execution, missing commit, output outsideallowedFiles).execution_failed: The tool or a required Git command failed to execute.timed_out: Execution exceeded the configured timeout.
Workflow statuses
completed: Every recorded stage completed successfully.completed_with_issues: Implementation ran, but one or more later stages did not complete successfully.failed: The workflow could not pass an early prerequisite such as planning.
Deprecation guidance
Structured JSON responses from the legacy alias run_gemini_coding_task
include a top-level machine-readable deprecation object:
"deprecation": {
"message": "run_gemini_coding_task is deprecated. Please use run_antigravity_coding_task instead.",
"replacement": "run_antigravity_coding_task"
}Workflow diagnostics
When run_development_workflow does not complete cleanly, it returns these
additional diagnostic fields:
failedStage: The name of the first failing stage (e.g."plan","implement","review","verify").nextAction: A descriptive recommendation on what to do next to resolve the issue.failureSummary: The failed stage summary lifted to the top level.failureDetails: The failed stage details lifted to the top level.
Workspace safety
By default, only the MCP process working directory and its descendants are allowed. Additional roots must be explicitly configured:
$env:SUPERPOWER_CODEX_ALLOWED_ROOTS = "C:\projects;D:\work"Use ; as the delimiter on Windows and : on macOS or Linux.
Additional safeguards:
Coding execution requires
allowExecution: true.Verification commands require
allowCommandExecution: true.Verification commands run without a shell; pipes, redirects, and command separators are rejected.
Strict execution compares reported files and commits with Git metadata.
allowedFilesaccepts only repository-relative paths.Antigravity does not recursively call the coding-task MCP tool.
Development
npm.cmd install
npm.cmd run typecheck
npm.cmd test
npm.cmd run buildThe verified baseline is 160 passing tests across 11 test files.
Updating an existing Windows installation
Download superpower-codex-mcp-v0.1.1-windows.zip from the GitHub release,
extract it, and run:
powershell -NoProfile -ExecutionPolicy Bypass `
-File .\scripts\update-installed.ps1 `
-InstallPath "C:\path\to\superpower-codex-mcp"The updater validates the installation, creates a backup under
.update-backups, installs production dependencies, and verifies MCP tool
discovery. Restart Codex Desktop after the update.
License
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Pushplayhero/superpower-codex-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server