MDPilot MCP
OfficialAllows using Google's Gemini 2.0 Flash model for generating markdown and AI instructions through the MCP server.
Allows using NVIDIA NIM's Llama 3.3 70B instruct model for free-tier generation via the MCP server.
Allows using OpenAI's GPT-4o model for generating markdown and AI instructions through the MCP server.
MDPilot
Markdown intelligence platform — generate production-grade AI instruction files, optimize tokens, and convert any file to markdown. Works in the browser and as an MCP server inside your IDE.
mdpilot.in · Docs · MCP setup · npm — mdpilot-mcp@0.3.1
What it generates
Mode | Input | Output |
Task (primary) | Ticket · Slack thread · spec | TASK.md + SPEC.md — structured, gap-checked, agent-ready |
Generate | 7-step wizard + tech stack | README · AGENTS · CLAUDE · CONTRIBUTING · SECURITY · SKILL · DESIGN · CONTEXT |
Convert | PDF · DOCX · CSV · HTML | Clean markdown via MarkItDown |
Explain | Code file or directory | WALKTHROUGH.md tuned to any audience |
Image → Prompt | Any image | Recreation prompt for FLUX / SD / Midjourney / DALL-E |
Interview primer | Role + level + JD | Ready-to-paste AI coach prompt |
Every output runs through a 5-pass token optimizer (20–40% reduction) and a self-verification loop that checks generated commands against your real repo before writing anything.
DESIGN.md follows the getdesign.md / awesome-design-md ecosystem spec — 9 sections in fixed order (Visual Theme → Agent Prompt Guide) so any design tool or AI agent that reads DESIGN.md files gets exact tokens, not guessed values.
Agent behavior directives (v0.3.1):
Human voice (
writingStyle: 'human') — natural prose for README, CONTRIBUTING, DESIGN; auto-enabled for non-technical and learner audiences; hard-ignored for agent files (AGENTS.md, CLAUDE.md, SKILL.md).Plan-first (ai_exec mode) — generated agent prompts now always open with "Before executing: write a 3-5 line plan" + step-verify gates before the Response style line.
Agent risk check (ai_exec + riskCheck toggle) — appends "check your plan against the Watch-outs above" to the agent prompt block when Watch-outs are present.
Related MCP server: Alcove
MCP server — use MDPilot from inside your IDE
The mdpilot-mcp npm package exposes all 10 tools directly to Claude Code, Cursor, Windsurf, and Goose. The server reads your actual repo on disk — generated files reference real scripts and paths, never guesses.
One-command setup (gets you a free key + writes the config):
npx -y mdpilot-mcp setupOr non-interactively:
npx -y mdpilot-mcp setup --key gsk_... --client claudeManual config (any MCP-compatible client):
{
"mcpServers": {
"mdpilot": {
"command": "npx",
"args": ["-y", "mdpilot-mcp"],
"env": { "GROQ_API_KEY": "gsk_..." }
}
}
}Accepted keys (tried in order — Groq and NVIDIA are free):
GROQ_API_KEY · NVIDIA_API_KEY · ANTHROPIC_API_KEY · OPENAI_API_KEY
10 MCP tools
Tool | What it does |
| Scan repo — detects stack, scripts, package manager, structure, MCP servers |
| Generate any instruction file grounded in real repo data |
| Turn a ticket or thread into a structured TASK.md |
| Generate WALKTHROUGH.md for any file or directory |
| Run the 5-pass optimizer (+ opt-in aggressive 6th pass) on any markdown |
| Analyze an image → recreation prompt |
| Detect stale docs — broken commands, broken paths, removed MCP servers |
| Patch only the stale sections, preserve everything else |
| Persist session state to CONTEXT.md (last 5 sessions, secret-redacted) |
| Load prior session context + annotate stale references inline |
Session memory quick-start — your data never leaves your machine:
# End of session:
"Use mdpilot to save our session context — decisions: X, next steps: Y"
# Start of next session:
"Load the project context via mdpilot"Multi-model
Provider | Model | Notes |
Anthropic | claude-sonnet-4-6 | Best instruction-file quality |
Groq | llama-3.3-70b-versatile | Free tier — recommended default |
NVIDIA NIM | meta/llama-3.3-70b-instruct | Free tier |
OpenAI | gpt-4o | |
gemini-2.0-flash |
Quick local setup
git clone https://github.com/get-mdpilot/MDPilot-mcp
cd MDPilot-mcp/mdpilot
npm install
cp .env.example .env.local # add at least one AI key
npm run dev # http://localhost:3000MCP server:
cd packages/mcp
npm install && npm run build
npm run inspect # MCP Inspector — test tools interactivelySee DEVELOPMENT.md for full architecture, file trees, environment variables, and contribution guide.
Stack
Next.js 16 · TypeScript · Tailwind CSS v4 · Supabase · @modelcontextprotocol/sdk · repomix · js-tiktoken · zod
Legal
Operated by Viveon Gizit Pvt Ltd · Bengaluru, India — Product Owner: Mohan
Privacy Policy · Terms of Service · Report an issue
MIT license
This server cannot be installed
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
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/get-mdpilot/MDPilot-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server