lumina-mcp
v1.3.2
Published
Unified MCP toolkit for engineering workflows, database connections, and source control
Downloads
1,437
Maintainers
Readme
Overview
Lumina MCP is a Model Context Protocol server that gives AI assistants (Antigravity IDE, Claude Code, Cursor, Codex, VS Code) direct access to your developer environment.
It acts as a secure bridge between your AI agent and the real tools developers use every day:
You (natural language) → AI Client (MCP) → Lumina MCP Server
↓
MySQL / PostgreSQL — GitHub — Jira / Trello / OpenProject
Schema inspection, Audit reports, Git commits, PR creation,
Ticket ingestion, Full dev cycle orchestration💡 Pro Tip: Lumina MCP works out of the box with any MCP-compatible AI client. You only need to include environment variables for the tools you actually plan to use.
Tools & Prompts
Database Tools (MySQL & PostgreSQL)
| Tool | MySQL | PostgreSQL | Description |
|------|:-----:|:----------:|-------------|
| execute_mysql_query / execute_postgres_query | ✅ | ✅ | Run read-only SQL queries via natural language |
| list_mysql_tables / list_postgresql_tables | ✅ | ✅ | List all tables in the target database |
| inspect_mysql_table / inspect_postgresql_table | ✅ | ✅ | Inspect columns, types, keys, nullability |
| analyze_mysql_query / analyze_postgresql_query | ✅ | ✅ | EXPLAIN/EXPLAIN ANALYZE with DBA-level verdict |
| save_audit_report / save_audit_report_pg | ✅ | ✅ | Persist AI-generated audit reports as Markdown |
Database Prompts:
| Prompt | Description |
|--------|-------------|
| running_query | Natural language → MySQL SELECT query with schema auto-detection |
| auditor_query | Full MySQL performance & security audit by a Principal DBA |
| running_pg_query | Natural language → PostgreSQL SELECT query with schema auto-detection |
| auditor_pg_query | Full PostgreSQL performance & security audit by a Principal DBA |
Example Prompts: In your MCP client, you can use these prompts as slash commands:
/running_query command="Run a query to find the top 5 customers with active status."
/auditor_query command="Analyze query performance for SELECT * FROM orders WHERE status = 'pending'"
/running_pg_query command="Show all tables in public schema."
/auditor_pg_query command="SELECT * FROM users"Version Control Tools (GitHub & Gitea)
| Tool | Description |
|------|-------------|
| generate_commit_and_push | Generate a Conventional Commit message, stage files, and push (provider-agnostic — works for GitHub, Gitea, or any git remote) |
| create_github_pr | Create a GitHub Pull Request with auto-generated tech company-style description |
| review_github_pr | Submit a rigorous AI code review directly to a GitHub Pull Request |
| fix_github_pr_review | Fetch PR review comments and automatically apply fixes to the codebase |
| get_github_pr_diff | Download a clean unified diff of any open GitHub PR for automated review |
| reply_to_pr_comment | Reply to inline comments within a GitHub Pull Request review |
| resolve_pr_review_thread | Resolve a GitHub PR review thread by its comment node ID |
Gitea (self-hosted)
Gitea reaches parity with GitHub for the four core capabilities — commit + push, create PR, fetch diff, and review PR. It is configured with a full base URL so any self-hosted instance (custom host, scheme, or port) is supported.
| Tool | Description |
|------|-------------|
| generate_commit_and_push | Shared with GitHub — commits and pushes to the current origin remote |
| create_gitea_pr | Create a Pull Request on a self-hosted Gitea instance |
| review_gitea_pr | Submit an AI code review (with optional inline comments) to a Gitea PR |
| fix_gitea_pr_review | Fetch Gitea PR review comments so the AI can apply fixes locally |
| get_gitea_pr_diff | Download a unified diff of a Gitea PR |
Configuration: set GITEA_BASE_URL (e.g. https://gitea.example.com:3000) and GITEA_TOKEN, or pass baseUrl directly to any Gitea tool. Auth uses a Gitea access token (Authorization: token <TOKEN>).
Known gaps vs GitHub (Gitea's API has no equivalent, so these tools are GitHub-only by design and are intentionally not registered for Gitea):
reply_to_pr_comment— Gitea has no 1:1 reply-to-inline-comment endpoint (its review-comment model differs).resolve_pr_review_thread— Gitea exposes no GraphQL API nor an equivalent "resolve review thread" semantic.
Git Prompts:
| Prompt | Description |
|--------|-------------|
| commit_generator_message | Conventional commit message generator with ticket/branch auto-detection |
| tech_company_pr_creator | Netflix/Meta/Google-style PR description generator |
| ai_code_reviewer | Senior Staff-level PR reviewer ([CRITICAL], [MAJOR], [MINOR], [NIT]) |
| fix_pr_review_message | Analyze review comments, apply fixes, and resolve threads automatically |
Example Prompts: In your MCP client, you can use these prompts as slash commands:
/commit_generator_message command="Stage my changes and generate a conventional commit message."
/tech_company_pr_creator command="Create a PR description from feat/auth into main."
/ai_code_reviewer command="Review PR #42 and approve it."
/fix_pr_review_message command="Resolve all review comments on PR #104."Project Management Integration
| Tool | Description |
|------|-------------|
| get_jira_ticket | Fetch a Jira issue by ID/Key — includes title, description, labels, comments, epic |
| get_trello_card | Fetch a Trello card — includes description, status, checklist, comment history |
| get_openproject_work_package | Fetch an OpenProject work package — includes assignee, priority, and comments |
| get_github_issue | Fetch a GitHub issue — includes body, comments, labels, milestones, and linked PRs |
| create_jira_ticket | Create a new Jira issue with labels, priority, and attachments |
| create_trello_card | Create a new Trello card in a specific list |
| create_openproject_work_package | Create a new OpenProject work package with assignee and priority |
| create_github_issue | Create a new GitHub issue with labels, milestone, and assignees |
PM Prompts:
| Prompt | Description |
|--------|-------------|
| pm_summarize_ticket | Summarize a raw Jira/Trello/OpenProject ticket as a Senior Product Manager |
| pm_brainstorm_plan | Brainstorm technical approach and create a step-by-step implementation plan |
| pm_test_catalog | Generate a comprehensive test catalog based on the ticket and technical plan |
| pm_create_ticket | Digest raw feature requests and structure them into Big Tech-standard tickets |
Example Prompts: In your MCP client, you can use these prompts as slash commands:
/pm_summarize_ticket command="Fetch Jira ticket LUM-402 and summarize it."
/pm_brainstorm_plan command="Download OpenProject work package #82 and create a technical plan."
/pm_test_catalog command="Get Trello card 64b19c and generate a test catalog."Orchestration
| Tool | Description |
|------|-------------|
| get_orchestration_phase | Retrieve system-level instructions for a specific orchestration phase (1–6) |
Orchestration Prompt:
| Prompt | Description |
|--------|-------------|
| lumina_orchestrate | Kicks off the full Orchestration Development Cycle. Has optional arguments: command and tokenBudget |
The orchestration cycle follows 6 deterministic phases:
| # | Phase | Description | |---|-------|-------------| | 1 | Discovery & Analysis | Fetches tickets, reads codebase, drafts requirements | | 2 | Technical Planning | Creates a precise file-by-file implementation plan | | 3 | Execution & Implementation | Implements code following the approved plan | | 4 | Iterative Code Review | Multi-agent review loop for security & quality | | 5 | Verification & Compounding | Runs tests, audits DB changes, records learnings | | 6 | Git Pull Request Release | Conventional commit, push, and GitHub PR creation |
Example Prompts: In your MCP client, you can use this prompt as a slash command:
/lumina_orchestrate command="Build a new login page" tokenBudget="save-tokens"
/lumina_orchestrate command="Implement requirements from Jira ticket LUM-402" tokenBudget="full-detail"Testing (Unit & E2E)
| Tool | Description | |------|-------------| | - | No standalone tools (driven via prompts) |
Testing Prompts:
| Prompt | Description |
|--------|-------------|
| create-unit-test | Senior SDET prompt to generate high-quality unit tests with >80% coverage. Covers Happy Path, Negative Path, Edge Cases, Security, Concurrency, and State across Frontend, Backend, and Mobile domains. |
| create-e2e-test | Senior SDET prompt to generate comprehensive end-to-end (E2E) tests. Covers Happy Path user journeys, Negative Path flows, Edge Cases, UI Stability, and Accessibility across any programming language/framework. |
Example Prompts: In your MCP client, you can use these prompts as slash commands:
/create-unit-test command="Write tests for src/auth/login.controller.ts"
/create-e2e-test command="Write e2e tests for the user registration flow."Security
All database connections are strictly read-only with multiple enforcement layers:
- Write blocking — Blocks
INSERT,UPDATE,DELETE,DROP,ALTER,CREATE,TRUNCATE - SQL injection detection — Detects comments (
--,/*), tautologies (OR 1=1), andUNION SELECT - Sensitive column filtering — Auto-strips
password,token,secret,credentialfrom results - Configurable restrictions — Extend blocked columns via
.lumina/database/restricColumn.json
Quick Start
Add this to your MCP client configuration (e.g. mcp.json, Cursor settings, Claude config):
{
"mcpServers": {
"lumina-mcp": {
"command": "npx",
"args": ["-y", "lumina-mcp"],
"env": {
"MYSQL_URL": "mysql://user:password@localhost:3306/db_name",
"POSTGRES_URL": "postgres://user:password@localhost:5432/db_name",
"GITHUB_TOKEN": "your-github-personal-access-token",
"GITEA_BASE_URL": "https://gitea.example.com:3000",
"GITEA_TOKEN": "your-gitea-access-token",
"JIRA_URL": "https://yourcompany.atlassian.net",
"JIRA_EMAIL": "[email protected]",
"JIRA_API_TOKEN": "your-jira-token",
"TRELLO_API_KEY": "your-trello-key",
"TRELLO_API_TOKEN": "your-trello-token",
"OPENPROJECT_URL": "https://openproject.yourcompany.com",
"OPENPROJECT_API_KEY": "your-openproject-key"
}
}
}
}You only need to include the environment variables for the integrations you use. If you already have the official Atlassian MCP or GitHub MCP installed, Lumina MCP will automatically use them as a fallback.
For detailed, client-specific installation guides (Antigravity IDE, Cursor, Claude Code, VS Code, Codex), visit Lumina MCP Documentation.
Documentation
Detailed guides for every tool, prompt, parameter, and usage example:
| Guide | Description | |-------|-------------| | MySQL Prompts & Tools | 5 MySQL tools, 2 prompts, security features, and examples | | PostgreSQL Prompts & Tools | 5 PostgreSQL tools, 2 prompts, security features, and examples | | GitHub Source Control | 7 Git/GitHub tools, 4 prompts, and fallback strategy | | Project Management Integration | 3 PM tools (Jira, Trello, OpenProject), 3 AI prompts | | AI Orchestration | 6-Phase Orchestration Engine, native fallback strategy | | Testing Module | Create unit tests with Big Tech QA standards |
Architecture
Lumina MCP follows a clean, layered architecture:
src/
├── index.ts # MCP server entry point
└── tools/
├── database/
│ ├── mysql/
│ │ ├── controller/ # Tool & prompt registration
│ │ ├── service/ # Business logic & read-only validation
│ │ └── repository/ # Connection pool
│ └── postgresql/
│ ├── controller/
│ ├── service/
│ └── repository/
├── github/ # Git & PR tools
├── projectmanagement/ # Jira, Trello, OpenProject tools
└── orchestration/ # Phase-based orchestration engineTesting
npm test # Run all tests
npm run test:coverage # Run tests with coverage report
npm run lint # Lint the codebase
npm run format # Format the codebaseContributing
We welcome contributions of all kinds! Please read CONTRIBUTING.md for guidelines on how to get started, coding standards, and the PR process.
