Dbt Docs MCP

🚀 DBT Documentation MCP Server

The Model Context Protocol (MCP) server is used to interact with dbt project metadata, including dbt Docs text assets (manifest.json, catalog.json). This server provides dbt graph information and allows queries for node details, model/column lineage, and related metadata.

✨ Features

This server offers the following tools:

• Search for dbt nodes:

  • Find nodes (models, sources, tests, etc.) by name (search_dbt_node_names).
  • Locate nodes by column name (search_dbt_column_names).
  • Search within the compiled SQL code of nodes (search_dbt_sql_code).

• View nodes:

  • Get detailed attributes for any given node's unique identifier (get_dbt_node_attributes).

• Explore lineage:

  • Find the direct upstream dependencies (predecessors) of a node (get_dbt_predecessors).
  • Find the direct downstream dependencies (successors) of a node (get_dbt_successors).

• Column-level lineage:

  • Trace all upstream sources of a specific column in a model (get_column_ancestors).
  • Trace all downstream dependencies of a specific column in a model (get_column_descendants).

• Suggested extensions:

  • A tool to execute SQL queries.
  • A tool to directly retrieve table/view/column metadata from the database.
  • A tool to search the knowledge base.

🚀 Quick Start

1. Prerequisites: Ensure that Python is installed and you have uv.
2. Clone the repository:

   git clone <repository-url>
   cd dbt-docs-mcp
   

3. Optional: Parse dbt documentation files for column-level lineage:
   • Set up the required Python environment, for example:

   uv sync
   

   • Use the provided script scripts/create_manifest_cl.py and provide the dbt documentation file, dbt directory, and the schema mapping and documentation CL files for the desired output path:

   python scripts/create_manifest_cl.py --manifest-path PATH_TO_YOUR_MANIFEST_FILE --catalog-path PATH_TO_YOUR_CATALOG_FILE --schema-mapping-path DESIRED_OUTPUT_PATH_FOR_SCHEMA_MAPPING --manifest-cl-path DESIRED_OUTPUT_PATH_FOR_MANIFEST_CL
   

   • Depending on the scale of your dbt project, generating column lineage may take some time (several hours).
4. Run the server:
   • If your target MCP client (Claude Desktop, Cursor, etc.) supports mcp.json, it should look like this:

   {
       "mcpServers": {
           "DBT Documentation MCP": {
           "command": "uv",
           "args": [
               "run",
               "--with",
               "networkx,mcp[cli],rapidfuzz,dbt-core,python-decouple,sqlglot,tqdm",
               "mcp",
               "run",
               "/Users/mattijs/repos/dbt-docs-mcp/src/mcp_server.py"
           ],
           "env": {
               "MANIFEST_PATH": "/Users/mattijs/repos/dbt-docs-mcp/inputs/manifest.json",
               "SCHEMA_MAPPING_PATH": "/Users/mattijs/repos/dbt-docs-mcp/outputs/schema_mapping.json",
               "MANIFEST_CL_PATH": "/Users/mattijs/repos/dbt-docs-mcp/outputs/manifest_column_lineage.json"
           }
           }
       }
   }