How to use custom or local embedding models (Ollama, vLLM) with mcp-server-qdrant?

4 answers

• Best answerAgentArmor teamTeam3d ago

  Still open: as of the last thread update (September 2026), the issue remains open, and neither of the two PRs mentioned in the thread has been merged — one is a draft and the other was closed without review. A maintainer confirmed the provider interface is flexible and invited PRs for a specific model provider, so this is a contribution path rather than an existing configuration option.

  Two PRs in the thread target the gap:

  • PR #187 (draft) adds an OpenAI-compatible embedding provider configured with the EMBEDDING_BASE_URL environment variable. Per the author, it covers any model behind an OpenAI-compatible /v1/embeddings endpoint, including local servers such as Ollama, vLLM, and LM Studio. It does not add custom FastEmbed model registration, so that part of the issue would remain open.
  • PR #143 adds Ollama as an embedding provider, following the existing FastEmbedProvider pattern, and supports local models such as bge-m3 and nomic-embed. It was closed after 60 days without review; the implementation is still available at the PR URL, and the author offered to reopen and rebase if there is renewed interest.

  A commenter proposed a config schema (proposal only, no implementation in the thread):

  jsonCopy to clipboard

  {
      "embedding": {
          "provider": "fastembed",
          "model": "BAAI/bge-small-en-v1.5"
      }
  }
  

  If you need custom embeddings today, the practical path is to watch or contribute to these PRs, or open a PR for the specific provider you need. As general advice, if you point EMBEDDING_BASE_URL at a remote service, note that your query text is sent to that endpoint.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @castlenthesky tried on GitHub (Outcome unclear)

  PR #187 (draft) adds an OpenAI-compatible embedding provider configured with the EMBEDDING_BASE_URL environment variable, covering models behind an OpenAI-compatible /v1/embeddings endpoint including Ollama, vLLM, and LM Studio; it does not cover custom FastEmbed model registration.

  Source: https://github.com/qdrant/mcp-server-qdrant/issues/62#issuecomment-5552143871 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @zsxh1990 tried on GitHub (Outcome unclear)

  PR #143 adds Ollama as an embedding provider following the existing FastEmbedProvider pattern, supporting local models like bge-m3 and nomic-embed.

  Source: https://github.com/qdrant/mcp-server-qdrant/issues/62#issuecomment-4861473602 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @magnus919 tried on GitHub (Didn't work)

  Reported that being limited to fastembed makes the server unusable for setups needing a custom model; the user's stack is Ollama with bge-m3:latest.

  Source: https://github.com/qdrant/mcp-server-qdrant/issues/62#issuecomment-2994965099 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport