Configuration
DeepPlan MCP is configured via environment variables in a .env file or your MCP server config.
Environment Variables
| Variable | Default | Description |
|---|---|---|
OPENROUTER_API_KEY | (required) | Your OpenRouter API key |
OPENROUTER_MODEL | anthropic/claude-sonnet-4 | Fallback model for single-call mode |
COUNCIL_MODEL | minimax/minimax-m2.5 | Model used for the 4 expert personas |
LEAD_ARCHITECT_MODEL | (same as COUNCIL_MODEL) | Model used for the Lead Architect synthesis |
DRAFT_PLAN_MAX_LENGTH | 8000 | Maximum characters for input draft plan |
Model Selection
Expert Personas (4 parallel calls)
The expert personas (Security, Performance, UX/DX, DevOps) use COUNCIL_MODEL. We recommend a fast, cost-effective model since these run in parallel:
- Recommended:
minimax/minimax-m2.5— fast, affordable, good analysis - Alternative:
google/gemini-2.0-flash— slightly more expensive, broader knowledge
Lead Architect (1 synthesis call)
The Lead Architect uses LEAD_ARCHITECT_MODEL. This should be a stronger model capable of synthesizing multiple reports:
- Recommended:
google/gemini-3-flash-preview— excellent synthesis, fast - Alternative:
anthropic/claude-sonnet-4— highest quality, slower
Tool Parameters
When calling upgrade_architecture_blueprint from your IDE:
| Parameter | Required | Description |
|---|---|---|
draft_plan | Yes | Your architecture plan text |
tech_stack | No | e.g. “SvelteKit, Cloudflare Workers, D1” |
model_choice | No | Override Lead Architect model for this call |
context_constraints | No | e.g. “Must run on Edge Runtime” |
Performance Tips
- Keep draft plans concise — 200-500 words is ideal
- Specify tech stack — enables stack-specific recommendations
- Add constraints — prevents incompatible suggestions