@okmira/mcp

MCP server and hooks for Mira — where stakeholders shape how AI builds your product.

What is Mira?

AI made developers 10x faster. Feedback is still 1x. Mira closes the gap.

Mira gives stakeholders a structured way to review builds and guide AI-assisted development — through feedback, reusable skills, and specialized agents. Developers fetch that input directly into their AI coding tool and apply it conversationally.

Installation

One-command setup (recommended)

npx @okmira/mcp setup

Opens your browser for authentication, detects your AI coding tool, and writes all config files automatically. Works with 18+ AI coding tools.

To refresh configs without re-authenticating (e.g. after a version update):

npx @okmira/mcp@latest setup --upgrade

Manual setup

Register the MCP server with your tool. Example for Claude Code:

claude mcp add mira \
  -e CC_FEEDBACK_API_URL=https://okmira.ai \
  -e CC_FEEDBACK_API_KEY=ccf_YOUR_KEY_HERE \
  -- npx @okmira/mcp

Get your API key from the project dashboard at okmira.ai.

MCP Tools

| Tool | Description | |------|-------------| | register_build | Register a new build after deployment. Use when auto-registration did not fire. | | fetch_feedback | Fetch feedback for a build. Returns role-contextualized feedback with suggested directions. | | update_feedback_status | Mark feedback as acknowledged, applied, or stale. Closes the feedback loop. | | check_pending_feedback | Check for unreviewed feedback across recent builds. Acts as an inbox. | | fetch_skills | List reusable skills (brand guidelines, code standards, etc.) defined by stakeholders. | | install_skill | Fetch a skill with full content for installation as a slash command. | | fetch_subagents | List specialized agent definitions (code reviewer, security auditor, etc.). | | install_subagent | Fetch a subagent definition with full content for installation. |

Hooks

Optional hooks for zero-friction automation. During setup, choose your deploy strategy:

| Strategy | How builds get registered | Hook behavior | |----------|--------------------------|---------------| | cli | Detects deploy URLs from CLI output | Matches platform CLI patterns (Vercel, Netlify, Cloudflare, etc.) | | git-push | Detects git push output | Uses deployUrl from .mira.json | | webhook | Vercel/external webhook POST | Skips PostToolUse hook entirely |

mira-auto-register (PostToolUse) — Detects deploys in Bash output and automatically registers builds. Strategy-aware: only fires for your configured workflow.

mira-check-feedback (SessionStart) — Surfaces pending feedback count at the start of each session so your AI tool can proactively inform you.

Hook credentials are read from the MCP server config — no duplication needed. Supported by Claude Code, Codex CLI, Factory.ai, and Kiro CLI.

Configuration

| Variable | Description | |----------|-------------| | CC_FEEDBACK_API_URL | Base URL — https://okmira.ai (production) or http://localhost:3000 (local) | | CC_FEEDBACK_API_KEY | Project-scoped API key (ccf_ prefix), generated from the dashboard |

How It Works

Developer deploys build
  → Build registered (via hook, MCP tool, or webhook)
  → Stakeholders review build, leave feedback and direction
  → Developer fetches feedback into their AI coding tool
  → AI applies feedback conversationally
  → New build deployed → cycle repeats

Requirements

• Node.js 18+ (for native fetch())
• git (for hooks — ref/branch detection)

License

MIT