Arcticdb MCP



Overview

arcticdb-mcp turns ArcticDB into an AI data copilot. Assistants can automate symbol updates, batch jobs, and time-series queries while built-in versioning protects every change. Go from question to data change in minutes, not scripts, with full traceability. You can also design your own workflows on top of arcticdb-mcp.

Table of Contents

• Demo Video
• Quickstart
• Installation
• Configuration
• Run Modes
• Tools
• Example Prompts
• Development
• Community Backlog
• Contributing
• License

Demo Video

Watch the live watch db demo:

https://github.com/user-attachments/assets/dd81f229-e1f9-46e6-8e79-7c7252bc25c6

Quickstart

1. Install and run with uvx (recommended):

uvx arcticdb-mcp

1. Configure your MCP client (Claude Desktop / Cursor / Windsurf / Continue):

{
  "mcpServers": {
    "arcticdb": {
      "command": "uvx",
      "args": ["arcticdb-mcp"],
      "env": {
        "ARCTICDB_URI": "lmdb:///path/to/your/database"
      }
    }
  }
}

1. Ask your assistant:

• "Show me the last 5 rows of AAPL in library finance"
• "Create a snapshot of finance before I update symbols"
• "List versions for symbol ES_intraday"

Installation

Choose one:

# Recommended
uvx arcticdb-mcp

# Or
pipx install arcticdb-mcp

# Or
pip install arcticdb-mcp

Python requirement: >=3.9

Configuration

Set ARCTICDB_URI in MCP client config or environment.

URI examples

• Local LMDB (Linux/macOS): lmdb:///path/to/db
• Local LMDB (Windows): lmdb://C:/path/to/db
• AWS S3: s3://s3.amazonaws.com:bucket?region=us-east-1&access=KEY&secret=SECRET
• Azure Blob: azure://AccountName=X;AccountKey=Y;Container=Z
• S3-compatible (MinIO, etc.): s3://your-endpoint:bucket?access=KEY&secret=SECRET

You can also use a .env file:

ARCTICDB_URI=lmdb:///path/to/db

Run Modes

stdio (default)

This is the default mode used by desktop MCP clients.

ARCTICDB_URI=lmdb:///path/to/db python -m arcticdb_mcp

HTTP / SSE

Set ARCTICDB_MCP_PORT to run over HTTP/SSE:

Use an integer from 1 to 65535. An unset or empty value selects stdio.

ARCTICDB_URI=lmdb:///path/to/db ARCTICDB_MCP_PORT=8000 python -m arcticdb_mcp

Endpoint:

http://localhost:8000/sse

Tools

Current server exposes 49 tools.

Arctic and Libraries

Symbols

Batch Operations

Snapshots

Query Helpers

Maintenance

Paginated Symbol Reads

Use read_symbol_page to inspect large symbols a page at a time without loading the full dataset into memory or overflowing assistant response context:

• Parameters: library, symbol, offset (default 0), limit (default 100, max 1000), optional columns, optional as_of.
• Backend Slicing: Row ranges and column projections are pushed directly to ArcticDB storage engines.
• Version Pinning: The initial page resolves the symbol's version (version). Pass as_of=version on subsequent pages to guarantee consistent reads across mutations.
• Continuation Info: Each response includes count, total_rows, has_more, and next_offset.

Two-Page Example

# Page 1: Read the first 100 rows of symbol AAPL
page1 = read_symbol_page(library="finance", symbol="AAPL", offset=0, limit=100)
# Returns: {"version": 2, "offset": 0, "limit": 100, "count": 100, "total_rows": 250, "has_more": True, "next_offset": 100, "data": [...]}

# Page 2: Pin version=2 and use next_offset to fetch the next slice
page2 = read_symbol_page(
    library="finance",
    symbol="AAPL",
    offset=page1["next_offset"],
    limit=100,
    as_of=page1["version"],
)
# Returns: {"version": 2, "offset": 100, "limit": 100, "count": 100, "total_rows": 250, "has_more": True, "next_offset": 200, "data": [...]}

Example Prompts

• "Read symbol NVDA from library finance"
• "Filter NVDA where price > 500 and volume >= 1000"
• "Resample ES_intraday to 5min and aggregate price mean, volume sum"
• "Write metadata owner=research to symbol ES_intraday"
• "Delete data in range for ES_intraday from 2024-01-01 to 2024-01-05"

Development

git clone https://github.com/YMuskrat/arcticdb-mcp
cd arcticdb-mcp
pip install -e .

Run locally:

ARCTICDB_URI=lmdb:///path/to/db python -m arcticdb_mcp

Test with MCP Inspector:

ARCTICDB_URI=lmdb:///tmp/test_db npx @modelcontextprotocol/inspector python -m arcticdb_mcp

Community Backlog

These ArcticDB capabilities are intentionally left open for contributors:

• stage
• finalize_staged_data
• sort_and_finalize_staged_data
• delete_staged_data
• get_staged_symbols
• read_batch_and_join
• admin_tools

Contributing

See CONTRIBUTING.md.

License

MIT