can't get qdrant embedding to work

[f1-outsourcing]

Do not really have much experience with using AI coding. Getting this error. Afaik qdrant only supports store and find.

Failed to connect to "qdrant-server" Error: fetch is not defined

name: mcp-qdrant
version: 0.0.1
schema: v1
mcpServers:
  - name: qdrant-server
    type: streamable-http
    url: http://192.168.x.x:8000/mcp/

[ibondarenko1]

The two things in your config are doing different jobs and the error is telling you which one is failing.

1. mcp-qdrant is MCP memory, not an embedding provider.

The MCP server you wired in (type: streamable-http, name mcp-qdrant) is the mcp-server-qdrant memory tool - it gives the model qdrant-store and qdrant-find tools so the model can read/write memory. It does not do embeddings for Continue's own RAG / autocomplete index. Continue's embedding configuration is a separate top-level block.

2. The fetch is not defined error is Continue running on Node 16 or older.

streamable-http MCP transport uses the global fetch() which is only available in Node 18+. Two fixes:

# verify
node --version  # need v18+

If you can't upgrade Node globally, switch to the stdio transport for the same MCP server (it doesn't need fetch):

mcpServers:
  - name: qdrant-server
    type: stdio
    command: uvx
    args: ["mcp-server-qdrant"]
    env:
      QDRANT_URL: "http://192.168.x.x:6333"
      COLLECTION_NAME: "memory"
      EMBEDDING_MODEL: "sentence-transformers/all-MiniLM-L6-v2"

uvx auto-installs the package so you don't need a separate venv.

3. If what you actually wanted was Continue using Qdrant for its codebase RAG, that is a different config block.

Continue's embeddingsProvider and docs indexer are separate. Add at top-level:

embeddingsProvider:
  provider: ollama
  model: nomic-embed-text
docs:
  - title: my-codebase
    startUrl: ./

Continue stores those embeddings in its local LanceDB by default; it does not currently use Qdrant as the codebase index backend (only as MCP memory). If you specifically need Qdrant as the codebase store, that's a feature request rather than a config issue.

4. Sanity check - what the MCP server actually does after you fix the transport.

Once mcp-qdrant connects, you should see two tools available to the agent:

qdrant-store(content, metadata)  # write a memory
qdrant-find(query)               # retrieve memory

These are agent-callable tools, not embedding endpoints. The model uses them to remember facts between sessions, not to index your code.

5. If you actually want a "Cursor-like" working setup with Continue + Ollama + Qdrant:

models:
  - name: codestral
    provider: ollama
    model: codestral:22b
    roles: [chat, edit, apply]
embeddingsProvider:
  provider: ollama
  model: nomic-embed-text
mcpServers:
  - name: qdrant-memory
    type: stdio
    command: uvx
    args: ["mcp-server-qdrant"]
    env:
      QDRANT_URL: "http://192.168.x.x:6333"
      COLLECTION_NAME: "continue-memory"

This gets you: Codestral on Ollama for completion + Continue's built-in indexer for codebase RAG + Qdrant as MCP memory for agent state.

Recipe: confirm Node 18+ -> use stdio transport for MCP -> understand MCP-qdrant is memory not embeddings -> add a separate embeddingsProvider block if you want codebase RAG -> verify both tools appear in the agent's tool list before testing the workflow.

[f1-outsourcing]

thanks for the detailed reply, upgrading worked indeed with getting the connections!

[omni-front]

You need to define a global fetch function for your environment to resolve the fetch is not defined error:

if (typeof fetch !== 'function') {
    global.fetch = require('node-fetch');
}

This ensures fetch is available in a Node.js environment, where it is not built-in by default.