DevPlan MCP Server
Transforms ideas into detailed, executable development plans with built-in verification, lessons learned tracking, and GitHub issue remediation workflows. Guides Claude through structured interviews, plan generation, execution with Haiku agents, and verification with Sonnet agents to maintain context and code quality across sessions.
README
DevPlan MCP Server
Transform ideas into executable development plans — an MCP server that brings the ClaudeCode-DevPlanBuilder methodology to Claude Code.
The Problem: AI coding assistants often lose context, skip steps, or produce inconsistent code across sessions.
The Solution: DevPlan creates detailed, Haiku-executable development plans with built-in verification, lessons learned, and issue remediation workflows.
flowchart LR
subgraph Planning["📋 Planning"]
A[Interview] --> B[Brief]
B --> C[Plan]
end
subgraph Execution["⚡ Execution"]
C --> D[Execute]
D --> E[Verify]
end
subgraph Learning["🧠 Learning"]
E -->|issues| F[Lessons]
F -->|improve| C
end
subgraph Remediation["🔧 Remediation"]
G[GitHub Issue] --> H[Parse]
H --> I[Task]
I --> D
end
style A fill:#e1f5fe,stroke:#0288d1
style F fill:#fff3e0,stroke:#f57c00
style G fill:#fce4ec,stroke:#c2185b
Key Features
| Feature | Description |
|---|---|
| Inline Methodology | All guidance is embedded — no external fetches needed |
| Haiku-Executable Plans | Plans so detailed that Claude Haiku can execute them |
| Lessons Learned | Captures issues from verification and injects them into future plans |
| Issue Remediation | Converts GitHub issues directly into remediation tasks |
| Tech Conflict Detection | Warns about incompatible technology choices |
| Executor & Verifier Agents | Auto-generates specialized agents for your project |
Install
claude mcp add devplan --transport sse https://devplan-mcp-server.mike-c63.workers.dev/sse
Or add to ~/.claude/mcp.json:
{
"mcpServers": {
"devplan": {
"type": "sse",
"url": "https://devplan-mcp-server.mike-c63.workers.dev/sse"
}
}
}
Quick Start
You: "Use devplan_start to help me build a CLI tool for managing dotfiles"
That's it. DevPlan will guide Claude through the entire process:
sequenceDiagram
participant You
participant Claude as Claude Code
participant MCP as DevPlan MCP
participant KV as Lessons KV
rect rgb(240, 248, 255)
Note over You,MCP: 📋 Planning Phase
You->>Claude: "build me a CLI tool"
Claude->>MCP: devplan_start()
MCP-->>Claude: inline methodology guidance
Claude->>You: Interview questions (one at a time)
You-->>Claude: Answers
Claude->>MCP: devplan_create_brief()
MCP-->>Claude: PROJECT_BRIEF.md
Claude->>MCP: devplan_generate_plan()
MCP->>KV: fetch lessons learned
KV-->>MCP: past lessons
MCP-->>Claude: DEVELOPMENT_PLAN.md + lessons
Claude->>MCP: devplan_generate_executor()
Claude->>MCP: devplan_generate_verifier()
end
rect rgb(255, 248, 240)
Note over You,Claude: ⚡ Execution Phase
loop Each Subtask
Claude->>Claude: Execute with Haiku agent
Claude->>Claude: Verify with Sonnet agent
Claude->>MCP: devplan_update_progress()
end
end
rect rgb(240, 255, 240)
Note over You,KV: 🧠 Learning Phase
Claude->>MCP: devplan_extract_lessons_from_report()
Claude->>MCP: devplan_add_lesson()
MCP->>KV: store for future projects
end
Claude-->>You: ✅ Project complete!
Usage Examples
New Project
"Use devplan_start to help me build [your idea]"
Fix a GitHub Issue
# Get issue JSON
gh issue view 123 --json number,title,body,labels,comments,url > issue.json
# Then tell Claude:
"Use devplan_issue_to_task with this issue to create a remediation plan"
Check Progress
"Use devplan_progress_summary to show me where we are"
Tools
Planning
| Tool | Purpose |
|---|---|
devplan_start |
Main entry point - guides Claude through the methodology |
devplan_interview_questions |
Get questions to gather project requirements |
devplan_create_brief |
Generate PROJECT_BRIEF.md |
devplan_parse_brief |
Parse existing brief into structured data |
devplan_list_templates |
List project templates (cli, web_app, api, library) |
Generation
| Tool | Purpose |
|---|---|
devplan_generate_plan |
Generate DEVELOPMENT_PLAN.md scaffold |
devplan_generate_claude_md |
Generate CLAUDE.md scaffold |
devplan_generate_executor |
Generate Haiku-powered executor agent |
devplan_generate_verifier |
Generate Sonnet-powered verifier agent |
Execution
| Tool | Purpose |
|---|---|
devplan_validate_plan |
Check plan completeness and structure |
devplan_get_subtask |
Get specific subtask details by ID |
devplan_update_progress |
Mark subtasks complete with notes |
devplan_progress_summary |
Get completion stats and next actions |
Lessons Learned
Feedback loop that captures issues from verification and incorporates them into future plans.
| Tool | Purpose |
|---|---|
devplan_add_lesson |
Capture a lesson from verifier findings |
devplan_list_lessons |
List accumulated lessons by severity |
devplan_archive_lesson |
Archive old lessons without deleting them |
devplan_delete_lesson |
Remove outdated or incorrect lessons |
devplan_extract_lessons_from_report |
Auto-extract lessons from verification reports |
Issue Remediation
Convert GitHub issues into structured remediation tasks — perfect for bug fixes and post-release maintenance.
| Tool | Purpose |
|---|---|
devplan_parse_issue |
Analyze a GitHub issue to extract requirements |
devplan_issue_to_task |
Generate remediation task with subtasks from an issue |
Analytics
| Tool | Purpose |
|---|---|
devplan_usage_stats |
View usage distribution across users (unique users, requests/user, histograms) |
flowchart LR
A["gh issue view 123 --json ..."] --> B[devplan_parse_issue]
B --> C{Analysis}
C --> D[Type: bug/feature]
C --> E[Severity: 🔴🟠🟡🔵]
C --> F[Components]
B --> G[devplan_issue_to_task]
G --> H[DEVELOPMENT_PLAN.md]
style A fill:#f5f5f5,stroke:#333
style H fill:#c8e6c9,stroke:#2e7d32
Architecture
graph TB
subgraph Client["Claude Code"]
CC[Claude Code CLI]
end
subgraph MCP["DevPlan MCP Server"]
SSE[SSE Endpoint]
Dash[Dashboard]
Tools[21 MCP Tools]
Gen[Plan Generators]
end
subgraph Storage["Cloudflare"]
KV[(KV Storage)]
DO[Durable Objects]
SQL[(SQLite per-DO)]
end
subgraph Methodology["Reference"]
GH[GitHub: ClaudeCode-DevPlanBuilder]
end
CC <-->|SSE| SSE
SSE --> Tools
Tools --> Gen
Gen --> KV
Tools --> DO
DO --> SQL
SQL -->|cleanup| KV
Dash --> KV
Gen -.->|examples| GH
style CC fill:#e3f2fd,stroke:#1565c0
style KV fill:#fff3e0,stroke:#ef6c00
style SQL fill:#e8f5e9,stroke:#388e3c
style GH fill:#f3e5f5,stroke:#7b1fa2
Storage Strategy
| Data | Location | Rationale |
|---|---|---|
| Session metadata | SQLite (per-DO) | Auto-deleted when DO destroyed |
| Lessons learned | KV | Global access across sessions |
| Aggregated analytics | KV | Fast dashboard reads |
| Cleanup schedules | DO Alarms | Native, reliable triggering |
Dashboard & Analytics
DevPlan includes a public dashboard for viewing aggregate usage statistics:
Dashboard URL: devplan-mcp-server.mike-c63.workers.dev/dashboard
API Endpoint: devplan-mcp-server.mike-c63.workers.dev/dashboard/api/stats
The dashboard shows:
- Summary cards: Total sessions, total tool calls, countries reached
- Line chart: Sessions and tool calls over the last 30 days
- Country table: Top 10 countries by session count
Session Tracking & Cleanup
Each MCP session is tracked with metadata stored in SQLite (per Durable Object):
- Session ID and timestamps (created, last activity)
- Geographic data (country, region, continent from Cloudflare)
- Tool call count and transport type (SSE/HTTP)
Automatic Cleanup: Sessions are automatically cleaned up to prevent storage bloat:
- After 7 days of inactivity, or
- After 30 days maximum age
When a session expires, its metrics are aggregated to KV storage before cleanup.
Privacy
All analytics are privacy-preserving:
- No IP storage: Only Cloudflare-derived country/region codes
- No user identification: Sessions are anonymous
- Auto-expiration: Daily stats expire after 90 days via KV TTL
- Public dashboard: Shows only aggregate statistics
Usage Stats via MCP
You can also query usage statistics programmatically:
"Use devplan_usage_stats to show me the last 7 days of usage"
This returns detailed breakdowns including unique users, requests per user, and distribution histograms.
Recent Updates
timeline
title DevPlan MCP Server - December 2025 / January 2026
section Week 6
Session Cleanup : Auto-cleanup after 7d inactivity or 30d max
Usage Dashboard : Chart.js visualization at /dashboard
Analytics API : JSON endpoint at /dashboard/api/stats
SQLite Tracking : Per-session metadata in Durable Objects
section Week 5
Usage Dashboard : Public dashboard with aggregate stats
Session Tracking : Track unique users per day
Analytics Tool : devplan_usage_stats MCP tool
section Week 4
Haiku-Executable Phases : Claude writes complete code, not scaffolds
Lesson Archiving : Archive old lessons without deletion
Severity Filtering : Filter lessons by min severity
Verifier Workflow : Prompts to run verifier at 100%
section Week 3
Content Drift Detection : Detects outdated inline guidance
Inline Methodology : No external fetches needed
Issue Remediation : GitHub issues → tasks
section Week 2
Error Recovery : Executor agent guidance
Lessons Enhancement : Active feedback loop
Verifier Agent : Auto-generate verifiers
section Week 1
Tech Conflict Detection : Warns on bad combos
Task Complete Sections : Squash merge workflow
Why DevPlan?
graph LR
subgraph Without["❌ Without DevPlan"]
A1[Vague prompt] --> A2[Inconsistent code]
A2 --> A3[Lost context]
A3 --> A4[Repeated mistakes]
end
subgraph With["✅ With DevPlan"]
B1[Structured interview] --> B2[Detailed plan]
B2 --> B3[Haiku executes]
B3 --> B4[Sonnet verifies]
B4 --> B5[Lessons captured]
B5 -.-> B2
end
style A4 fill:#ffcdd2,stroke:#c62828
style B5 fill:#c8e6c9,stroke:#2e7d32
| Without DevPlan | With DevPlan |
|---|---|
| Context lost between sessions | Plans preserve full context |
| Inconsistent code quality | Haiku follows exact specifications |
| Same mistakes repeated | Lessons learned system prevents recurrence |
| No verification step | Sonnet actively tries to break the code |
| Bugs found in production | Issues caught before release |
Development
npm install
npm run dev # Local development
npm run deploy # Deploy to Cloudflare Workers
Contributing
Contributions welcome! Please see the ClaudeCode-DevPlanBuilder repo for methodology details.
Community
Love DevPlan? Share your experience with #devplanmcp on social media!
License
MIT
<p align="center"> <b>Built for Claude Code</b><br> <a href="https://modelcontextprotocol.io">Model Context Protocol</a> • <a href="https://workers.cloudflare.com/">Cloudflare Workers</a> • <a href="https://github.com/mmorris35/ClaudeCode-DevPlanBuilder">DevPlanBuilder Methodology</a> <br><br> 💬 <b>#devplanmcp</b> </p>
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.