Bug in chroma_list_collections causes individual collection names to appear concatenated to Claude

[andreicozma1]

See example:

● Let me list all current collections to see the exact formatting:
  ⎿  collection1    
  ⎿  mycollection
  ⎿  global
  ⎿  user
  ⎿  main

● I can see the issue! The collection names appear as a single continuous string without any separators on my end:
  collection1mycollectionglobalusermain

  This makes it impossible for me to distinguish individual collection names.

This causes agents using this MCP server to fail to discover and query collections on their own,

[nealrauhauser]

This is indeed a bug that I had to fix in my fork of this code. I used Claude Code to do it, so I don't know the details, but a diff of the original server.py vs. the one in my code base would easily reveal the fix.

[jairad26]

strange, i think this might be an issue with the mcp client rather than the server? looking at the response to the list_collections tool, it returns an array of the individual names as elements. and looking at the snippet above, it looks like the client was not able to correctly parse an array

chroma-mcp/src/chroma_mcp/server.py

Line 166 in e19e4b3

return [coll.name for coll in colls]

there is also a test to verify the tool call returns an array, which also seems to work

chroma-mcp/tests/test_server.py

Line 533 in e19e4b3

async def test_list_collections_success():

if you can get a repro of the tool returning a single continuous string (i was not able to), id be happy to try and fix it.

What mcp client are you using? i tried repro'ing using claude desktop as well and wasn't able to

[svperfecta]

I can confirm this is a problem. Repro steps for me:

1. Done outside of MCP: Initialize two collections, add documents to each collection, use default embeddings. Essentially this:

            for i, chunk in enumerate(chunks):
                chunk_id = f"{base_id}_chunk_{i}" if len(chunks) > 1 else base_id
                ids.append(chunk_id)
                documents.append(chunk)
                metadatas.append({
                    "filename": md_file.name,
                    "filepath": str(md_file),
                    "chunk_index": i,
                    "total_chunks": len(chunks)
                })
            
            # Upsert to ChromaDB (idempotent operation)
            collection.upsert(
                ids=ids,
                documents=documents,
                metadatas=metadatas
            )

2. Open up Claude Code and:

Can you show me all of the collections you know about in Chroma
  ⎿  reference
  ⎿  services

⏺ There is only one collection in your ChromaDB: referenceservices