SQLite-Vec MCP

Overview

SQLite-Vec MCP is a Model Context Protocol server built on top of SQLite and the sqlite-vec extension, providing vector database capabilities directly within SQLite. Released in July 2026 with over 980 GitHub stars, this server enables AI agents to perform vector similarity search, hybrid keyword+vector queries, and automated schema inference—all within a single-file, zero-dependency database.

This MCP server is particularly well-suited for edge deployments and resource-constrained environments where running separate vector databases (Pinecone, Qdrant, Chroma) is impractical. With SQLite-Vec, agents get vector search capabilities with the simplicity and portability of SQLite.

Features

• Vector Index Creation: Create and manage vector indexes directly in SQLite
• Similarity Search: Support for L2 distance, cosine similarity, and dot product
• Hybrid Search: Combine keyword and vector searches in a single query
• Zero External Dependencies: Single SQLite file with sqlite-vec extension
• Auto Schema Inference: Automatically infer schema from JSON data and create vector tables
• Incremental Indexing: Add vectors to existing indexes without rebuilding
• Full-Text Search Integration: Combine with SQLite FTS5 for powerful hybrid queries
• Multiple Vector Dimensions: Support for arbitrary vector dimensions

Installation

# Install via pip (Python MCP server)
pip install mcp-sqlite-vec

# Or via npm (Node.js MCP server)
npm install -g @mcp-sqlite-vec/server

# Launch via npx (no installation)
npx @mcp-sqlite-vec/server --database ./vectors.db

Configuration

{
  "mcpServers": {
    "sqlite-vec": {
      "command": "npx",
      "args": [
        "@mcp-sqlite-vec/server",
        "--database", "./vectors.db"
      ],
      "env": {
        "SQLITE_VEC_MODEL": "all-MiniLM-L6-v2"
      }
    }
  }
}

Environment Variables:

Variable	Description	Default
SQLITE_VEC_MODEL	Embedding model for vector generation	all-MiniLM-L6-v2
SQLITE_VEC_DIMENSIONS	Vector dimensions	384
MCP_LOG_LEVEL	Logging verbosity	info

Available Tools

Tool	Description
create_vector_table	Create a table with vector columns
insert_vectors	Insert documents with their embeddings
vector_search	Perform similarity search on vectors
hybrid_search	Combine keyword and vector search
auto_schema_inference	Infer schema from JSON data and create vector table
list_tables	List all vector tables in the database
table_info	Get information about a specific vector table
delete_vectors	Remove vectors from a table

Usage Examples

Example 1: Create Vector Table and Insert Documents

# Via MCP client
mcp_client.call_tool("sqlite-vec", "create_vector_table", {
    "table_name": "documents",
    "columns": {
        "id": "INTEGER PRIMARY KEY",
        "content": "TEXT",
        "embedding": "VECTOR(384)",
        "metadata": "TEXT"
    }
})

mcp_client.call_tool("sqlite-vec", "insert_vectors", {
    "table_name": "documents",
    "documents": [
        {"id": 1, "content": "AI agents are transforming software development", "metadata": '{"category": "tech"}'},
        {"id": 2, "content": "Vector databases enable semantic search", "metadata": '{"category": "data"}'},
        {"id": 3, "content": "SQLite is perfect for embedded applications", "metadata": '{"category": "database"}'}
    ]
})

Example 2: Vector Similarity Search

response = mcp_client.call_tool("sqlite-vec", "vector_search", {
    "table_name": "documents",
    "query": "What are AI agents?",
    "distance_metric": "cosine",  # "l2", "cosine", "dot_product"
    "top_k": 5
})

# Results sorted by similarity

Example 3: Hybrid Keyword + Vector Search

response = mcp_client.call_tool("sqlite-vec", "hybrid_search", {
    "table_name": "documents",
    "query": "AI agents development",
    "keyword_fields": ["content"],
    "vector_field": "embedding",
    "top_k": 10,
    "weight_vector": 0.7,  # Vector search weight
    "weight_keyword": 0.3  # Keyword search weight
})

Example 4: Auto Schema Inference

response = mcp_client.call_tool("sqlite-vec", "auto_schema_inference", {
    "json_file": "./documents.json",
    "table_name": "auto_documents",
    "vector_field": "content",
    "metadata_fields": ["title", "author", "date"]
})

Example 5: List Vector Tables

response = mcp_client.call_tool("sqlite-vec", "list_tables", {})
# Returns: {"tables": ["documents", "auto_documents", "embeddings"]}

Advanced Features

Custom Embedding Models

# Use different embedding models
mcp_client.call_tool("sqlite-vec", "set_embedding_model", {
    "model": "text-embedding-ada-002",
    "dimensions": 1536,
    "api_key": "sk-xxx"
})

Batch Processing

# Insert large batches efficiently
mcp_client.call_tool("sqlite-vec", "insert_vectors", {
    "table_name": "documents",
    "json_file": "./large_corpus.json",
    "batch_size": 1000
})

Index Optimization

# Rebuild vector index for better performance
mcp_client.call_tool("sqlite-vec", "rebuild_index", {
    "table_name": "documents"
})

Edge Deployment

SQLite-Vec MCP is ideal for edge deployments:

# Deploy on Raspberry Pi or embedded devices
import sqlite3
from miniagent7b import MiniAgent

# SQLite-Vec requires only ~100MB total
agent = MiniAgent(model_path="miniagent-7b-int4", device="cpu")
vector_db = sqlite3.connect("./vectors.db")

Pros

• ✅ Zero Dependencies: Single SQLite file, no external services
• ✅ Vector Search Native: Built-in vector similarity search
• ✅ Hybrid Search: Combine keyword and vector queries
• ✅ Auto Schema Inference: Automatically create vector tables from JSON
• ✅ Edge-Ready: Perfect for resource-constrained deployments
• ✅ SQLite Familiarity: Leverages existing SQLite knowledge

Cons

• ✅ SQLite Limitations: Not suitable for high-concurrency scenarios
• ✅ Embedding Quality: Depends on external embedding model choice
• ✅ New Server: Limited production track record
• ✅ Scaling Limits: SQLite has inherent scalability constraints

When to Use

• Edge Deployments: Deploy vector search on devices without cloud access
• Prototyping: Quick experimentation without infrastructure setup
• Local Agent Workflows: Embed vector search in local agent applications
• Resource-Constrained Environments: Limited memory or compute availability
• Hybrid Search Needs: Combine keyword and semantic search capabilities
• Single-File Databases: Need portable, backupable database solutions

Resources

• GitHub Repository
• SQLite-Vec Documentation
• MCP Protocol