Troubleshooting
Common issues and how to fix them.
Neo4j Connection Refused
If you see ServiceUnavailable: Unable to connect to bolt://localhost:7687:
- Make sure Neo4j is running:
docker ps | grep neo4j - Start it if needed:
docker-compose up -d neo4j - Verify the port isn't blocked by a firewall.
Embedding API Key Errors
If indexing fails with an API key error:
- Set
GEMINI_API_KEYorGOOGLE_API_KEY - If using OpenAI for code embeddings, set
CODE_EMBEDDING_PROVIDER=openaiandOPENAI_API_KEY
MCP Server Not Found by Clients
If Claude or Cursor can't connect:
- Verify the server is running:
agentic-memory serve - Check if port 8000 is in use:
lsof -i :8000 - Try a different port:
agentic-memory serve --port 3000 - Validate your client config JSON syntax.
Semantic Search Returns Poor Results
If search results don't match your query:
- Verify your embedding API key is set correctly.
- Check that chunks exist in Neo4j:
MATCH (ch:Chunk) RETURN count(ch) - Re-index if you recently changed the embedding provider.
Out of Memory During Indexing
For large codebases:
- Increase Neo4j heap size:
NEO4J_dbms_memory_heap_max__size=4G - Reduce indexed file extensions in
.codememory/config.json - Use
agentic-memory watchfor incremental updates instead of full re-indexes.