feat: add Ollama embedding provider for local models

[zsxh1990]

What

Adds Ollama as an embedding provider alongside FastEmbed, enabling local embedding models without external API keys.

Why

From #62: "Being stuck with fastembed only means this mcp server won't work at all for a lot of us. I'm using Ollama/bge-m3:latest" — @magnus919

@kacperlukawski confirmed: "We have a pretty flexible interface, so adding a new embedding model should not be a big deal. Please feel free to open a PR."

How

New file: src/mcp_server_qdrant/embeddings/ollama.py

class OllamaProvider(EmbeddingProvider):
    """Ollama implementation using /api/embed endpoint."""
    
    def __init__(self, model_name: str, base_url: str = "http://localhost:11434"):
        ...
    
    async def embed_documents(self, documents: list[str]) -> list[list[float]]:
        return await self._embed(documents, input_type="search_document")
    
    async def embed_query(self, query: str) -> list[float]:
        return (await self._embed([query], input_type="search_query"))[0]

Updated files:

• types.py — Add OLLAMA = "ollama" to EmbeddingProviderType
• factory.py — Add OLLAMA case to create_embedding_provider()
• settings.py — Add ollama_base_url field (env: OLLAMA_BASE_URL)

Usage

# .env
EMBEDDING_PROVIDER=ollama
EMBEDDING_MODEL=bge-m3
OLLAMA_BASE_URL=http://localhost:11434

Or with any Ollama-compatible model:

EMBEDDING_PROVIDER=ollama
EMBEDDING_MODEL=nomic-embed-text

Testing

# Start Ollama
ollama serve
ollama pull bge-m3

# Run MCP server
EMBEDDING_PROVIDER=ollama EMBEDDING_MODEL=bge-m3 python -m mcp_server_qdrant

Design decisions

1. Uses /api/embed endpoint — Ollama's native embedding API, supports batch embedding
2. Auto-detects vector size — Embeds a dummy text on first call to get dimensions
3. Separate base_url setting — Allows connecting to remote Ollama instances
4. Async with httpx — Non-blocking I/O, consistent with FastEmbed provider

Ref

Closes #62

[zsxh1990]

Hi @kacperlukawski — friendly check-in on this PR 🙂 It's been about 4 weeks since the Ollama embedding provider was added. The implementation follows the existing pattern and enables local embedding models (like ) without external API keys, as requested in #62.

Let me know if there's anything that needs adjusting — happy to rebase or make changes as needed!

[zsxh1990]

Hi @kacperlukawski — second friendly check-in on this PR 🙂

Status:

• 28 days since open (4 weeks)
• 1 day since first check-in, no maintainer response
• PR mergeable=True, no conflicts, base=master
• No equivalent Ollama embedding PR has been merged (checked Add Ollama embedding provider support #108/feat: add OpenAI-compatible embedding provider #111/Add OpenAI-compatible embedding provider support #92 — all OpenAI-compatible or closed)

If the maintainer team is happy with the direction, I am happy to:

• Rebase to main if master → main migration is planned
• Add tests if needed
• Adjust naming/API to match qdrant-client conventions

If there is no bandwidth to review right now, no problem — happy to keep it open as a reference. If closing is preferred, I can close it cleanly with a summary of what was implemented for future contributors.

Thanks for your time.

[zsxh1990]

Hi @kacperlukawski — third check-in on this PR 🙂

Status update:

• 30 days since open (4+ weeks)
• 2 previous check-ins, no maintainer response
• PR is mergeable, no conflicts, base=master
• All CI checks pass

The Ollama embedding provider enables local model usage without API keys — a common request from the community. Would love to get this merged or receive feedback. Thanks!

[zsxh1990]

Hi @qdrant/mcp-server-qdrant maintainers! 👋

Just a friendly bump on this PR. It's been open for about 5 weeks.

Summary: Adds Ollama as an embedding provider for local model support. This enables users to run embeddings locally without cloud API dependencies.

Changes:

• New class
• Support for any Ollama-compatible model
• Auto-detection of available models

Is there anything I can do to help move this forward? Happy to address any feedback or make adjustments.

Thanks! 🙏

[zsxh1990]

Closing this PR as it has been open for 60 days with no review decision.

This was a contribution to add Ollama embedding provider support for local models. The PR includes tests and documentation. If there's renewed interest, happy to reopen with updated code.

Thanks for maintaining mcp-server-qdrant!