dbn-go-mcp-data

command

v0.8.11 Latest Latest Go to latest Published: Mar 20, 2026 License: Apache-2.0 Imports: 8 Imported by: 0

Details

Valid go.mod file
Redistributable license
Tagged version
Stable version
Learn more about best practices

Repository

github.com/WT-In/dbn-go

Links

Open Source Insights

README ¶

`dbn-go-mcp-data` Databento MCP Server with DuckDB Cache

dbn-go-mcp-data is a full-featured Model Context Protocol (MCP) server for Databento services. It includes all the metadata discovery tools from dbn-go-mcp-meta, plus data download (fetch_range) and a local DuckDB-backed Parquet cache for efficient re-queries.

CAUTION: The fetch_range tool incurs Databento billing! All other tools are free.

It requires your Databento API Key. Since this program is typically executed by a host program (like Claude Desktop), the preferred method is to store the API key in a file and use --key-file or the DATABENTO_API_KEY_FILE environment variable. It supports both STDIO and SSE MCP transports.

$ dbn-go-mcp-data --help
usage: dbn-go-mcp-data -k <api_key> [opts]

  -h, --help              Show help
  -k, --key string        Databento API key (or set 'DATABENTO_API_KEY' envvar)
  -f, --key-file string   File to read Databento API key from (or set 'DATABENTO_API_KEY_FILE' envvar)
  -l, --log-file string   Log file destination (or MCP_LOG_FILE envvar). Default is stderr
  -j, --log-json          Log in JSON (default is plaintext)
  -c, --max-cost float    Max cost, in dollars, for a query (<=0 is unlimited) (default 1)
      --cache-dir string  Directory for cached data files (default "~/.dbn-go/cache/")
      --cache-db string   Path to DuckDB database file (default "~/.dbn-go/cache.duckdb")
      --read-only         Disable fetch_range tool (no billing possible)
  -p, --hostport string   host:port to listen to SSE connections (default "127.0.0.1:8889")
      --sse               Use SSE Transport (default is STDIO transport)
  -v, --verbose           Verbose logging

Available Tools

Discovery Tools (no billing)

These are identical to those in dbn-go-mcp-meta:

Tool	Params	Description
`list_datasets`	`start`?, `end`?	Lists all available Databento dataset codes, optionally filtered by date range.
`list_schemas`	`dataset`	Lists all available data record schemas for a given dataset.
`list_fields`	`schema`	Lists all field names and types for a given schema (JSON encoding).
`list_publishers`	`dataset`?	Lists all publishers with their publisher_id, dataset, venue, and description.
`get_dataset_range`	`dataset`	Returns the available date range (start and end) for a dataset.
`get_dataset_condition`	`dataset`, `start`?, `end`?	Returns data quality/availability condition per day.
`list_unit_prices`	`dataset`	Lists unit prices in USD per gigabyte for each schema and feed mode.
`resolve_symbols`	`dataset`, `symbols`, `start`, `stype_in`?, `stype_out`?, `end`?	Resolves symbols from one symbology type to another.
`get_cost`	`dataset`, `schema`, `symbols`, `start`, `end`, `stype_in`?	Returns estimated cost in USD, data size, and record count for a query.

Data Tools

Tool	Params	Description
`fetch_range`	`dataset`, `schema`, `symbols`, `start`, `end`, `stype_in`?, `stype_out`?	Fetches market data and caches as Parquet. Returns metadata (view name, record count, size). Incurs billing.

Supported schemas for fetch_range: ohlcv-1s, ohlcv-1m, ohlcv-1h, ohlcv-1d, trades, mbp-1, tbbo, imbalance, statistics.

Cache Tools (no billing)

Tool	Params	Description
`query_cache`	`sql`	Query cached data with DuckDB SQL. Returns results as CSV.
`list_cache`	(none)	Lists cached datasets with view names, file counts, and sizes.
`clear_cache`	`dataset`?, `schema`?	Removes cached data, optionally filtered by dataset/schema.

The recommended workflow for an LLM is: list_datasets -> list_schemas -> list_fields -> get_cost -> fetch_range -> query_cache.

Read-Only Mode

Use --read-only to disable the fetch_range tool entirely. In this mode, the server only exposes metadata discovery tools and cache tools — no new data can be fetched and no Databento billing can be incurred. This is useful for environments where you want to query previously cached data without risk of accidental charges.

dbn-go-mcp-data --read-only --key-file /path/to/databento_api_key.txt

Metadata tools (list_datasets, get_cost, etc.) still require an API key and remain fully functional. Cache tools (query_cache, list_cache, clear_cache) work as normal against any previously fetched data.

How Caching Works

Data fetched via fetch_range is stored as Parquet files under ~/.dbn-go/cache/{dataset}/{schema}/. An in-memory DuckDB instance creates views backed by these files via read_parquet(). The LLM discovers available views using list_cache and queries them with standard SQL through query_cache.

MCP Host Configuration

Claude Desktop

Add to your Claude Desktop configuration file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "dbn": {
      "command": "dbn-go-mcp-data",
      "args": [
        "--key-file", "/path/to/databento_api_key.txt",
        "--max-cost", "1.50"
      ]
    }
  }
}

Claude Code

claude mcp add --transport stdio dbn -- dbn-go-mcp-data --key-file /path/to/databento_api_key.txt

SSE Mode

For networked deployments, run over HTTP with Server-Sent Events:

dbn-go-mcp-data --sse                                  # localhost only (default)
dbn-go-mcp-data --sse --hostport 0.0.0.0:8889          # listen on all interfaces