# Openclaw Github MCP Install

> Install, configure, verify, and troubleshoot the official GitHub MCP server for an OpenClaw bot, including auth identity checks, permissions, tool-loading verification, and write-action guardrails.

- Skill: `mikepans1013/openclaw-github-mcp-install` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add mikepans1013/openclaw-github-mcp-install`
- Raw SKILL.md: https://api.skillmd.com/api/skills/mikepans1013/openclaw-github-mcp-install/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: mikepans1013 (https://skillmd.com/u/mikepans1013)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/mikepans1013/openclaw-github-mcp-install

---


# OpenClaw GitHub MCP Install

Use this skill when an OpenClaw bot needs only GitHub MCP setup or troubleshooting. This skill intentionally excludes Google OAuth and other onboarding steps.

## Operating Modes

- **Prep only** — inspect, explain, and prepare, but do not save secrets or modify live MCP config.
- **Live install** — install/configure GitHub MCP now.
- **Repair** — diagnose and repair an existing GitHub MCP install.
- **Rollback** — remove live MCP config and move live artifacts to recoverable trash.

If the request is ambiguous, default to prep-only until the user clearly authorizes live install or repair.

## First Response Contract

Before making changes, explicitly confirm:

1. Prep-only, live install, repair, or rollback?
2. Read-only or write-capable?
3. Public-only, selected private repos, or broad private repo access?
4. Is local secret storage under `~/.openclaw/secrets/` authorized on this host?

## First Questions

Before configuring anything, ask:

1. Which human/operator owns this bot?
2. Which GitHub user or organization should this bot connect to?
3. Should access be public-only, selected private repos, or broad private repo access?
4. Should the bot be allowed to perform write actions after explicit approval, or read-only only?
5. Is this a fresh GitHub MCP install or a repair of an existing one?
6. Should the bot only prepare the install plan, or perform the live install now?

Never assume the current `gh` login, SSH key, local git remote, or another bot's token is the intended GitHub identity.

## Safety Rules

- Use GitHub's official MCP server as the source of truth: https://github.com/github/github-mcp-server
- Never paste, print, commit, or summarize GitHub tokens, PATs, OAuth tokens, private keys, or generated configs containing secrets.
- Store credentials via OpenClaw/Gateway secret handling when available, or under `~/.openclaw/secrets/` with `chmod 600`.
- Verify the authenticated GitHub identity before using repo access.
- Do not delete repos, change visibility, force-push, merge PRs, publish releases, change repo settings, or perform destructive writes without explicit approval for the exact action.

## Files in This Skill

- `references/github-mcp-openclaw.md` — OpenClaw-specific setup, verification, and troubleshooting notes.
- `references/install-nuances.md` — specific pitfalls encountered while setting up GitHub access for OpenClaw bots.

Read `references/github-mcp-openclaw.md` before installing or repairing GitHub MCP. Read `references/install-nuances.md` when troubleshooting auth or tool-loading problems.

## Standard Workflow

1. Confirm target GitHub account/org and desired repo scope.
2. Confirm whether the task is prep-only, live install, repair, or rollback.
2. Open the official GitHub MCP server docs and follow the current install/auth path.
3. For OpenClaw-managed MCP config, prefer `openclaw mcp set` / `openclaw mcp unset`.
4. Configure the MCP server in OpenClaw using a stable server name such as `github` or `github-official`.
4. Reload/restart only as required by OpenClaw's current MCP configuration path.
5. Verify MCP tools load.
6. Verify the authenticated GitHub identity and expected repo visibility.
7. Run a harmless read operation.
8. If writes are needed, test on a safe test repo/branch before using production repos.
9. Document only safe facts in the bot's `TOOLS.md`.

## Install Profiles

Prefer one of these explicit profiles:

- **read-only**
- **write-capable with explicit per-action approval**
- **repair existing install**
- **rollback existing install**

## Final Report

Report:

- MCP server name
- GitHub account/org connected
- Access scope verified
- Tool-loading status
- Harmless read test result
- Any blocked permissions or missing scopes
- Safety boundaries for future writes
- Whether the result is prep-only, live, repaired, or rolled back