Weaviate MCP: collection not found (fix the class name and capitalization)

Not specified by the original author.

TL;DR: The Weaviate instance is fine; the collection name is wrong. List the collections that exist, fix the name (Weaviate class names start uppercase by convention), and retry.

Collection "documents" not found

Fix it

1. List the collections on the instance. Use the MCP server's schema/list tool, or:

curl -H "your auth header https://your-cluster.weaviate.cloud/v1/schema | python3 -m json.tool | grep '"class"'

Expected: the real class names, e.g. Documents, Articles.

2. Fix the name in your query. Watch for:

   • Capitalization: Documents vs documents. Weaviate convention is uppercase-first.
   • Plurals and prefixes you assumed.
   • The collection living on a different instance (dev vs prod).

3. Retry the tool call.

   Expected: results instead of not-found.

When to use this

• The server connects, but queries fail with collection/class not found.
• The name was typed by hand or guessed by the agent.

When NOT to use this

• Connection or 401 errors. Fix reachability and auth first.
• No collections exist at all. Then import or create the schema first.

Compatibility

• weaviate/mcp-server-weaviate.
• Weaviate 1.2x+.

Why it happens

Weaviate does not fuzzy-match class names. Agents reconstruct names from conversation context and get the casing or pluralization slightly wrong. The error is exact because the lookup is exact.

Edge cases

• Multi-tenancy: the collection exists but the tenant does not. Tenant names are exact too.
• If the collection was just created, schema propagation is fast but not instant. Wait a few seconds on large clusters.
• Deleted collections stay gone. Recreate the schema and re-import if it was dropped.

Weaviate organizes data into collections (classes), and querying a nonexistent collection fails. Weaviate class names must start with an uppercase letter by convention, and the name is exact. MCP agents frequently guess collection names or miscapitalize them. The fix is listing existing collections through the MCP server's schema tools or the REST /v1/schema endpoint and correcting the name.