Architectural Intelligence Engine for AI-Powered Development
Turn vague plans into production-ready architecture blueprints —
reviewed by a Council of AI Architects before you write a single line of code.
Every AI coding assistant can generate code — but architecture is where projects succeed or fail. DeepPlan catches design flaws, security gaps, and performance bottlenecks before they become expensive rewrites.
| Without DeepPlan | With DeepPlan | |
|---|---|---|
| Planning | AI gives generic, shallow plans | 4 expert architects analyze in parallel |
| Quality | No guardrails — hallucination drift | Constitution enforces architectural correctness |
| Perspective | Single viewpoint | Security + Performance + UX/DX + DevOps |
| Output | Vague suggestions | Numbered, executable steps for your AI agent |
Your Draft Plan → 🔒 Security · ⚡ Performance · 🎨 UX/DX · 🔧 DevOps → 👑 Lead Architect → Blueprint
DeepPlan uses the Model Context Protocol (MCP) — it works with any MCP-compatible client, not just IDEs:
| Client | Use Case | |
|---|---|---|
| 🖥️ | Cursor, Windsurf, VS Code Copilot | Architecture review while coding |
| 💻 | Claude Code | Terminal-based planning and brainstorming |
| 🤖 | Claude Desktop, ChatGPT | Conversational architecture consulting |
| 🔧 | Custom MCP Clients | Build your own AI-powered architecture tools |
| 📡 | CI/CD & Automation | Automated architecture linting in pipelines |
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"deepplan": {
"command": "npx",
"args": ["-y", "deepplan-mcp"],
"env": {
"DEEPPLAN_API_KEY": "dpk_your-key-here"
}
}
}
}Add to your Windsurf MCP configuration:
{
"mcpServers": {
"deepplan": {
"command": "npx",
"args": ["-y", "deepplan-mcp"],
"env": {
"DEEPPLAN_API_KEY": "dpk_your-key-here"
}
}
}
}claude mcp add deepplan -- npx -y deepplan-mcp --api-key=dpk_your-key-hereAdd to .vscode/mcp.json:
{
"servers": {
"deepplan": {
"command": "npx",
"args": ["-y", "deepplan-mcp"],
"env": {
"DEEPPLAN_API_KEY": "dpk_your-key-here"
}
}
}
}Get your API key at deepplan.dev/api-keys — free tier included.
"Build me a SaaS with Stripe billing and multi-tenant auth"
Get a complete architecture blueprint — data model, API design, auth flow, deployment strategy — reviewed by 4 experts before writing a single line of code.
"Migrate our REST API to GraphQL"
Get expert analysis of migration risks, breaking changes, edge cases, and a step-by-step execution plan your AI coding assistant can follow.
"Review this authentication flow for vulnerabilities"
Get a Security Architect's threat model: injection vectors, data exposure risks, and concrete mitigations tied to your tech stack.
"Optimize this data pipeline that's hitting 5s response times"
Get bottleneck analysis with specific optimizations: caching strategy, query patterns, cold start reduction — all for your actual stack.
"Design a public REST API for our platform"
Get DX analysis, versioning strategy, error handling patterns, rate limiting architecture, and documentation structure.
"Automatically review architecture decisions in pull requests"
Integrate DeepPlan into your pipeline to catch architectural anti-patterns before they reach production.
Send a draft plan to the Council of Architects for expert-level review.
| Parameter | Required | Description |
|---|---|---|
draft_plan |
Yes | The architecture plan text to upgrade |
tech_stack |
No | e.g. "SvelteKit, Cloudflare Workers, D1" |
context_constraints |
No | e.g. "Must run on Edge Runtime" |
model_choice |
No | AI model for the Lead Architect synthesis |
Returns: A structured blueprint with Architecture Directives, Edge Cases, Constraints, Patterns, and numbered Next Steps your AI agent can execute immediately.
Analyze a plan and recommend which expert personas are most relevant.
| Parameter | Required | Description |
|---|---|---|
draft_plan |
Yes | The plan to analyze |
available_personas |
No | Custom persona list (defaults to 4 built-in experts) |
Every council session enforces architectural guardrails called "The Constitution" — rules that prevent AI hallucination drift and ensure high-quality output:
- Terminology Precision — Flags misused architectural terms (e.g., "fetch" for a DB query → should be "query")
- Clean Architecture Metrics — Enforces low coupling, high cohesion, correct dependency direction
- No Code Output — Experts focus on WHAT and WHY, never HOW in code
- Tech Stack Awareness — All recommendations must reference your actual stack
- Impact Prioritization — Most critical findings first, no filler
| Persona | Focus Areas |
|---|---|
| 🔒 Security Architect | Auth, data exposure, injection, CORS, API abuse, secrets management |
| ⚡ Performance Engineer | Latency, caching, token cost, cold starts, DB queries, concurrency |
| 🎨 UX/DX Designer | Integration smoothness, error clarity, onboarding, config simplicity |
| 🔧 DevOps Engineer | Deploy strategy, CI/CD, monitoring, disaster recovery, scaling |
| 👑 Lead Architect | Synthesizes all expert reports into a unified, executable blueprint |
Want domain-specific experts? Create custom personas at deepplan.dev — available via proxy mode.
Uses your DeepPlan API key (dpk_...). Requests are analyzed by DeepPlan Scout (expert analysis) and synthesized by DeepPlan Architect (blueprint generation).
DEEPPLAN_API_KEY=dpk_your-key-here
Premium features via proxy mode:
- Custom personas and councils
- Advanced orchestration (debate mode, AI+IDE refinement)
- AI-powered search context (Librarian)
- Usage tracking and credit management
Uses your own OpenRouter API key. The council runs locally with 4 built-in expert personas. No DeepPlan account needed.
OPENROUTER_API_KEY=sk-or-v1-your-key-here
| Variable | Description |
|---|---|
DEEPPLAN_API_KEY |
DeepPlan API key — enables proxy mode with full features |
OPENROUTER_API_KEY |
OpenRouter key — enables direct mode (fallback) |
DRAFT_PLAN_MAX_LENGTH |
Max input characters (default: 8000) |
Configuration priority: CLI args > environment variables > .env file
npx deepplan-mcp --api-key=dpk_your-key-hereDeepPlan is designed to be extended. The open-source core gives you building blocks to create your own architectural intelligence:
- Custom Personas — Add domain-specific experts (Database Architect, ML Engineer, etc.) in
src/prompts/personas.ts - Search Providers — Plug in your own search backend by implementing the
SearchProviderinterface insrc/search/types.ts - Custom Constitutions — Define your team's architectural standards and coding conventions
- Pipeline Integration — Use as a pre-commit hook, CI/CD step, or automated PR reviewer
- Build Your Own Client — The MCP protocol means any client can connect — build custom tools on top of DeepPlan
┌─────────────────┐ stdio ┌──────────────┐
│ MCP Client │◄────────────►│ DeepPlan │
│ │ MCP │ MCP Server │
│ • AI IDEs │ Protocol │ (this repo) │
│ • CLI Tools │ └──────┬───────┘
│ • Custom Apps │ │
└─────────────────┘ ┌─────────────┴──────────────┐
│ │
Proxy Mode Direct Mode
│ │
┌──────▼──────┐ ┌────────▼────────┐
│ DeepPlan API │ │ OpenRouter │
│ deepplan.dev │ │ API │
│ │ │ │
│ • Scout │ │ • 4 built-in │
│ (analysis) │ │ personas │
│ • Architect │ │ • Basic council │
│ (synthesis)│ │ │
│ • Debate │ │ │
│ • AI Search │ │ │
└──────────────┘ └──────────────────┘
The open-source client provides full council functionality with 4 built-in personas. For teams and power users, deepplan.dev offers:
| Feature | Open Source | DeepPlan Pro |
|---|---|---|
| 4 Built-in Experts | ✅ | ✅ |
| Custom Personas | — | ✅ |
| Council Builder | — | ✅ |
| Debate Mode (Pro/Con) | — | ✅ |
| AI Search (Librarian) | — | ✅ |
| Chair Persona | — | ✅ |
| Constitution Editor | — | ✅ |
| Usage Dashboard | — | ✅ |
# Clone
git clone https://github.com/gapgapweiqi/deepplan-mcp.git
cd deepplan-mcp
# Install
npm install
# Dev (with hot reload)
npm run dev
# Build
npm run build
# Type check
npm run typecheckWe welcome contributions! See CONTRIBUTING.md for guidelines.
MIT — use it however you want.
Built with ❤️ by DeepPlan