Auto-start ChromaDB on MCP Server Startup - Implementation Summary

Servers Affected

• ai/mcp/server/memory-core/
• ai/mcp/server/knowledge-base/

Changes Implemented

1. ChromaManager.mjs (both servers)

• Changed connect() to return boolean instead of throwing

    async connect() {
      try {
          await this.client.heartbeat();
          this.connected = true;
          return true;
      } catch (e) {
          this.connected = false;
          logger.debug('[ChromaManager] ChromaDB not accessible:', e.message);
          return false;  // Don't throw - graceful degradation
      }
  }

2. DatabaseLifecycleService.mjs (both servers)

• Added initAsync() method for auto-start

    async initAsync() {
      await super.initAsync();
      await ChromaManager.ready();     // Initialize ChromaManager
      await this.startDatabase();      // Auto-start ChromaDB if needed
  }

• Auto-start respects existing ChromaDB instances (service-managed or external)
• No changes needed to startDatabase() - already handles all scenarios

3. SessionService.mjs (memory-core only)

• Removed duplicate ChromaDB client - now uses ChromaManager singleton
• Removed dbClient property from construct()
• Added dependency wait in initAsync():

    async initAsync() {
      await super.initAsync();
      await DatabaseLifecycleService.ready();  // Ensure ChromaDB available
      
      // Use ChromaManager instead of direct client
      this.memoryCollection   = await ChromaManager.getMemoryCollection();
      this.sessionsCollection = await ChromaManager.getSummaryCollection();
  }

4. mcp-stdio.mjs (both servers)

• Changed startup sequence:

    // Before
  await ChromaManager.ready();
  
  // After
  await DatabaseLifecycleService.ready();  // Auto-starts ChromaDB

• Removed race condition workaround comment (no longer needed)

5. config.mjs

knowledge-base:

• Fixed ChromaDB data path:

    // Before
  path: path.resolve(process.cwd(), 'dist/ai-knowledge-base.jsonl'),  // ❌ Wrong - JSONL file
  
  // After
  path: path.resolve(process.cwd(), 'chroma'),  // ✅ ChromaDB data directory
  dataPath: path.resolve(process.cwd(), 'dist/ai-knowledge-base.jsonl'),  // ✅ JSONL path renamed

• Set debug: true (temporary, will be reverted)

memory-core:

• Cleaned up path: './chroma-memory' → 'chroma-memory' (removed unnecessary ./)
• Set debug: true (temporary, will be reverted)

6. logger.mjs (all 3 servers)

• Added logger.log method to github-workflow, knowledge-base, and memory-core:

    logger.log = createLogMethod('log');

7. DatabaseService.mjs (knowledge-base only)

• Updated to use new config property:

    // In createKnowledgeBase() and embedKnowledgeBase()
  const outputPath = aiConfig.dataPath;  // Was: aiConfig.path

8. DatabaseLifecycleService.mjs (memory-core only)

• Added debug logging:

    logger.error('Starting ChromaDB (Memory Core) process...');

• Removed Observable import (now imported globally in mcp-stdio.mjs)

9. mcp-stdio.mjs (memory-core only)

• Added Observable import at top level:

    import Observable from '../../../../src/core/Observable.mjs';

  This ensures the Observable mixin is loaded globally, so services using static observable = true don't need to import it individually
• Updated comment: Removed "fixes race condition" since initialization order is now explicit

Benefits Achieved

• ✅ ChromaDB automatically starts on server initialization
• ✅ Respects external/existing ChromaDB instances (no duplicates)
• ✅ Services properly wait for dependencies via .ready() pattern
• ✅ Single source of truth for ChromaDB access (ChromaManager)
• ✅ Graceful degradation preserved (server starts even if auto-start fails)
• ✅ Fixed config bug in knowledge-base (wrong path type)

Verification

Both servers now start successfully without manual ChromaDB startup:

• memory-core: Port 8001, 205 memories, 16 summaries, auto-summarization works
• knowledge-base: Port 8000, 0 documents (empty but healthy)