Qdrant MCP: Not existing vector name error (named vs unnamed vector mismatch)

Not specified by the original author.

TL;DR: Not existing vector name means the collection's vector layout does not match what the MCP server expects. Named vs unnamed vectors are getting mixed up. Check the collection's vector config and align the server's expected vector name with it.

Not existing vector name error

Fix it

1. Inspect the collection's vector configuration:

curl -H "api-key value YOUR_KEY" your-cluster/collections/your-collection | python3 -m json.tool | grep -A 10 vectors

Look for whether vectors is a single config (unnamed) or a map of names (named vectors).

2. Compare with what the MCP server is configured to use. The server settings (e.g. vector name, embedding provider output) must match the collection.

3. Fix the mismatch one of two ways:

   • Point the server at the collection's actual vector name, or
   • Recreate the collection with the layout the server expects, then re-index.

4. Restart the MCP server.

   Expected: store and search operations succeed.

When to use this

• The collection exists and is reachable, but store/search fail with the vector name error.
• The collection was created by a different tool than the MCP server.

When NOT to use this

• The error is collection not found. That is a name problem, not a vector-layout problem.
• Connection or auth errors. Fix those first.

Compatibility

• qdrant/mcp-server-qdrant and compatible forks with configurable vector names.
• Qdrant collections with named vectors (multi-vector setups).

Why it happens

Qdrant supports both unnamed vectors (one per point) and named vectors (several per point, e.g. dense plus sparse). The MCP server was configured against one layout, but the collection was created with the other, often by a different tool or an earlier server version. The names do not line up, so every vector operation misses.

Edge cases

• Recreating the collection deletes its data. Export or re-index afterward.
• Embedding dimension must also match. A 1536-dim vector into a 768-dim collection fails differently but is worth checking at the same time.
• If multiple tools share the collection, agree on one vector naming scheme across all of them.

The mcp-qdrant-custom troubleshooting guide (cedriccmh/mcp-qdrant-custom docs/TROUBLESHOOTING.md) documents this exact issue: Vector Type Mismatch with symptom Error: Not existing vector name error, store operation fails, collection exists but search fails. The root cause is a mismatch between the collection vector type and the server configuration: the collection uses unnamed vectors while the server expects named vectors, or vice versa. The diagnosis is inspecting the collection's vectors config and aligning the server configuration. Source: https://github.com/cedriccmh/mcp-qdrant-custom/blob/HEAD/docs/TROUBLESHOOTING.md

Not specified by the original author.

TL;DR: Your Weaviate instance does not have the vectorizer module your collection needs. Enable it with ENABLE_MODULES at container startup (e.g. text2vec-openai), provide its API key, and recreate the collection.

vectorizer: module "text2vec-openai" not enabled

Fix it

1. Check which modules the instance runs:

curl https://your-cluster.weaviate.cloud/v1/meta | python3 -m json.tool

Look under modules.

2. For Docker, enable the module at startup:

docker run -d --name weaviate -p 8080:8080 -p 50051:50051 \
  -e ENABLE_MODULES=text2vec-openai \
  -e OPENAI_APIKEY [your value] \
  semitechnologies/weaviate:latest

3. Collections created before the module was enabled will not magically gain vectors. Recreate the collection (or re-import) after enabling.

4. Restart the MCP client.

   Expected: vector search works.

When to use this

• The error names a vectorizer module as not enabled or not found.
• Vector search fails on a fresh self-hosted instance while keyword search works.

When NOT to use this

• Weaviate Cloud. Modules there are managed; pick a supported vectorizer in the collection config instead.
• The error is about the API key for the vectorizer (invalid key). Enablement is fine; the key is wrong.

Compatibility

• weaviate/mcp-server-weaviate against self-hosted Weaviate.
• text2vec-openai, text2vec-cohere, text2vec-transformers modules.

Why it happens

Weaviate core does not vectorize text by itself; modules do. The Docker image enables only what ENABLE_MODULES lists, defaulting to none. Collections reference a vectorizer by name, and at query time Weaviate discovers the module is absent.

Edge cases

• text2vec-transformers runs a local model and needs significant RAM. Prefer an API-based vectorizer on small machines.
• Changing the vectorizer on an existing collection requires re-importing all objects. There is no in-place migration.
• Module names are versioned in newer Weaviate. Check the docs for the exact module name for your version.

Weaviate vectorizes text through modules (text2vec-openai, text2vec-cohere, text2vec-transformers) that must be enabled at instance startup via ENABLE_MODULES. A Weaviate instance started without the needed module rejects vectorizer-dependent operations. The weaviate-cloud secure coding rules show collections configured with vectorizer settings, and the standard fix for missing-vectorizer errors is enabling the module in the Docker environment or choosing a vectorizer the instance actually runs.

Not specified by the original author.

TL;DR: Verify the index exists with the exact name before debugging query vectors: a wrong name gives empty results, not an error. Always put $vectorSearch first in the pipeline, then $project, $limit, and post-stages after it. After $vectorSearch, use score with the $meta vectorSearchScore projection to rank or threshold results.

Fix

1. Verify the index exists with the exact name before debugging query vectors: a wrong name gives empty results, not an error. Expected: the value or setting is exactly right, nothing ambiguous.

2. After $vectorSearch, use score with the $meta vectorSearchScore projection to rank or threshold results. Expected: this specific failure stops.

Details

Always put $vectorSearch first in the pipeline, then $project, $limit, and post-stages after it. Context: Vector search reference notes (grounded in the official MongoDB docs): the $vectorSearch stage MUST be the first stage in an aggregation pipeline; anything before it errors or is ignored. The index field must name a real vector search index on the collection, and if the name is misspelled or does not exist, MongoDB returns no results rather than an error. Requirements: an Atlas cluster on a supported version and a vector search index with vector-type fields on the collection. Agents that compose $vectorSearch mid-pipeline or typo the index name get silent empty results and blame the embeddings.

When to use

You hit exactly this: vectorSearch must be the first pipeline stage; a wrong index name returns empty results in vectorSearch.

When not to use

A different error, or the same symptom in a different tool. This page only covers the failure above.

Compatibility

vectorSearch.

Source: Vector search reference notes (grounded in the official MongoDB docs): the $vectorSearch stage MUST be the first stage in an aggregation pipeline; anything before it errors or is ignored. The index field must name a real vector search index on the collection, and if the name is misspelled or does not exist, MongoDB returns no results rather than an error. Requirements: an Atlas cluster on a supported version and a vector search index with vector-type fields on the collection. Agents that compose $vectorSearch mid-pipeline or typo the index name get silent empty results and blame the embeddings.

Verified resolution: Always put $vectorSearch first in the pipeline, then $project, $limit, and post-stages after it. Verify the index exists with the exact name before debugging query vectors: a wrong name gives empty results, not an error. After $vectorSearch, use score with the $meta vectorSearchScore projection to rank or threshold results.

Source URL: https://github.com/fcakyon/claude-codex-settings/blob/HEAD/plugins/mongodb-skills/skills/mongodb-search-and-ai/references/vector-search.md