Project Memory MCP

#!/usr/bin/env node import { Server } from '@modelcontextprotocol/sdk/server/index.js'; import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js'; import { CallToolRequestSchema, ListToolsRequestSchema, } from '@modelcontextprotocol/sdk/types.js'; import { INIT_PROMPT } from './prompts/init-prompt.js'; import { ORGANIZE_PROMPT } from './prompts/organize-prompt.js'; import { composePrompt, getProjectRoot } from './utils/prompt-loader.js'; import { TASK_JSON_SCHEMA } from './schemas/task-schema.js'; /** * Project Memory MCP Server * * A pure prompt provider that returns instructions for Claude to execute. * Never reads/writes project files directly - only loads prompt templates. */ const server = new Server( { name: 'project-memory-mcp', version: '0.1.0', }, { capabilities: { tools: {}, }, } ); /** * List available tools */ server.setRequestHandler(ListToolsRequestSchema, async () => { return { tools: [ { name: 'init', description: 'Initialize project memory system. Creates folder structure, generates project-specific prompts, and sets up claude.md instructions. Only run once per project.', inputSchema: { type: 'object', properties: {}, }, }, { name: 'parse-tasks', description: 'Parse tasks from spec files or implementation plans. Extracts tasks with IDs, descriptions, acceptance criteria, dependencies, and adds them to tasks-active.json after user approval.', inputSchema: { type: 'object', properties: {}, }, }, { name: 'review', description: 'Review uncommitted code changes. Analyzes git diff, checks against current tasks and architecture, identifies issues, and proposes task/architecture updates for user approval.', inputSchema: { type: 'object', properties: {}, }, }, { name: 'sync', description: 'Sync project memory with recent commits. Updates tasks (marks completed), prunes commit log to last 20 commits, updates architecture if needed, and extracts new commands.', inputSchema: { type: 'object', properties: {}, }, }, { name: 'organize', description: 'Organize existing CLAUDE.md into project-memory structure. Migrates architecture, conventions, commands, tasks, and specs from CLAUDE.md to .project-memory/ files while keeping minimal references. Requires user approval.', inputSchema: { type: 'object', properties: {}, }, }, ], }; }); /** * Handle tool calls - returns prompts only, never executes operations */ server.setRequestHandler(CallToolRequestSchema, async (request) => { const { name } = request.params; const projectRoot = getProjectRoot(); try { let prompt: string; switch (name) { case 'init': // Only hardcoded prompt in the system prompt = INIT_PROMPT; break; case 'parse-tasks': prompt = await composePrompt( projectRoot, 'parse-tasks.md', `# Task Parsing You are helping parse tasks from a spec or implementation plan. ## Task Schema ${TASK_JSON_SCHEMA} ## Instructions 1. Read the spec file from .project-memory/specs/ or from the user's message 2. Extract tasks with unique IDs (TASK-001, TASK-002, etc.) 3. Include: title, description, acceptance criteria, dependencies, priority, subtasks if needed 4. Check existing tasks in .project-memory/tasks/tasks-active.json to avoid duplicates 5. Show parsed tasks to user via AskUserQuestion for approval 6. After approval, update tasks-active.json using Write or Edit tool 7. Set specReference field to the spec file path Remember: Get user approval before writing any files. `.trim() ); break; case 'review': prompt = await composePrompt( projectRoot, 'review.md', `# Code Review You are helping review uncommitted code changes. ## Task Schema ${TASK_JSON_SCHEMA} ## Instructions 1. Get git diff using Bash tool: \`git diff\` and \`git diff --cached\` 2. Read current context: - .project-memory/tasks/tasks-active.json - .project-memory/architecture.md - .project-memory/specs/*.md (if relevant) 3. Analyze changes for: - Code quality issues - Potential bugs or security issues - Alignment with architecture - Task progress 4. Propose updates via AskUserQuestion: - Task status changes - Architecture updates - Issues found 5. After approval, apply changes using Write/Edit tools Remember: Get user approval before writing any files. `.trim() ); break; case 'sync': prompt = await composePrompt( projectRoot, 'sync.md', `# Post-Commit Sync You are helping sync project memory with recent commits. ## Task Schema ${TASK_JSON_SCHEMA} ## Instructions 1. Get commit history using Bash: \`git log --oneline -20\` 2. Read current state: - .project-memory/tasks/tasks-active.json - .project-memory/tasks/tasks-completed.json - .project-memory/commit-log.md - .project-memory/architecture.md 3. Determine task completions based on commits 4. Propose updates via AskUserQuestion: - Move completed tasks to tasks-completed.json - Update commit-log.md (keep last 20 commits) - Update architecture.md if structure changed - Add new commands to useful-commands.md 5. After approval, apply changes using Write/Edit tools Remember: Get user approval before writing any files. `.trim() ); break; case 'organize': // Organize/migrate existing CLAUDE.md into project-memory prompt = ORGANIZE_PROMPT; break; default: throw new Error(`Unknown tool: ${name}`); } return { content: [ { type: 'text', text: prompt, }, ], }; } catch (error) { const errorMessage = error instanceof Error ? error.message : 'Unknown error occurred'; return { content: [ { type: 'text', text: `Error: ${errorMessage}`, }, ], isError: true, }; } }); /** * Start the server */ async function main() { const transport = new StdioServerTransport(); await server.connect(transport); console.error('Project Memory MCP server running on stdio'); } main().catch((error) => { console.error('Fatal error:', error); process.exit(1); });