Adf Graph

MCP server that builds a queryable dependency graph from Azure Data Factory pipeline artifacts.

Parses ADF pipelines, datasets, linked services, and SQL stored procedures into a graph of nodes and edges. Exposes MCP tools for dependency analysis, impact assessment, data lineage tracing, and environment/overlay management.

Quick Start

npm install
npm run build
ADF_ROOT=/path/to/work-repo node dist/server.js

ADF_ROOT must point to a directory containing pipeline/, dataset/, and/or linkedService/ subdirectories. The graph is built lazily on the first tool call and rebuilt automatically when files change.

Multi-environment Configuration

To query multiple ADF roots (e.g. your work repo and a live dev factory export), create adf-graph.json in the repo root (next to dist/):

{
  "environments": {
    "work-repo": {
      "path": "C:/path/to/work-repo",
      "default": true
    },
    "dev1": {
      "path": "C:/path/to/adf-export/dev1"
    }
  }
}

Each environment has its own lazily-built graph and staleness tracker. Use the optional environment parameter on any tool to target a specific environment. Omit it to use the default.

You can also set ADF_CONFIG=/absolute/path/to/adf-graph.json to use a config file at an arbitrary location (takes priority over the sidecar file).

Overlays

Environments can have an optional overlays array to layer local or in-progress files on top of the base graph:

{
  "environments": {
    "work-repo": {
      "path": "C:/repos/adf-main",
      "default": true,
      "overlays": ["C:/my-wip/", "C:/one-off/NewPipeline.json"]
    }
  }
}

• Overlay directories with ADF structure (pipeline/, dataset/) are parsed normally.
• Loose files are auto-detected by inspecting JSON content.
• The base graph stays clean; a merged view appears as {name}+overlays.
• Runtime overlays can be added/removed via MCP tools (graph_add_overlay, graph_remove_overlay).
• Runtime additions are ephemeral (lost on server restart).

MCP Configuration

Single environment (ADF_ROOT)

{
  "mcpServers": {
    "adf-graph": {
      "command": "node",
      "args": ["/path/to/adf-graph/dist/server.js"],
      "env": {
        "ADF_ROOT": "/path/to/work-repo"
      }
    }
  }
}

Multi-environment (config file)

{
  "mcpServers": {
    "adf-graph": {
      "command": "node",
      "args": ["/path/to/adf-graph/dist/server.js"],
      "env": {
        "ADF_CONFIG": "/path/to/adf-graph.json"
      }
    }
  }
}

Tools

All tools except graph_list_environments accept an optional environment parameter. Omit it to use the default environment.

Tool	Description
graph_stats	Node/edge counts by type, build time, staleness flag
graph_find_consumers	All pipeline activities that read/write/call a given artifact
graph_describe_pipeline	Pipeline summary, activity list, or full detail with SQL queries and column mappings
graph_impact_analysis	All nodes affected upstream/downstream if an artifact changes
graph_data_lineage	Data flow paths for a Dataverse entity or table (resolves unqualified names like Org_Staging to dbo.Org_Staging); includes source tables from pipeline parameters; optional column filter and depth limit
graph_find_paths	All dependency paths between two named nodes
graph_search_queries	Search across all activity SQL and FetchXML for a text pattern
graph_diff_pipeline	Compare a pipeline's structure across two environments
graph_list_environments	List all configured environments with paths, stats, and staleness status
graph_add_overlay	Add an overlay path to an environment (runtime, ephemeral)
graph_remove_overlay	Remove a runtime overlay from an environment
graph_list_overlays	List all overlays (config + runtime) for an environment
graph_add_environment	Register a new ephemeral environment pointing to an ADF root
graph_remove_environment	Remove a runtime environment

graph_find_consumers / graph_impact_analysis target types

pipeline, activity, dataset, stored_procedure, table, dataverse_entity

graph_describe_pipeline depth values

• summary — name, parameters, child pipelines, root orchestrators
• activities — adds activity list with dependencies, sources, and sinks
• full — adds SQL queries, FetchXML, and column-level mapping details

graph_data_lineage depth limiting

Use the optional maxDepth parameter to limit traversal depth (number of hops). Useful for large graphs where unlimited traversal produces oversized results.

graph_search_queries

Searches all activity sqlReaderQuery and FetchXML fields for a case-insensitive substring match. Returns matching activities with the full query text.

graph_diff_pipeline

Compares a pipeline's structure across two named environments. Reports added, removed, and modified activities with details on what changed (SQL, column mappings, sources, sinks).

Directory Structure

src/
  config.ts          # Config loader (ADF_CONFIG / adf-graph.json / ADF_ROOT)
  server.ts          # MCP server entry point
  graph/
    model.ts         # Graph, NodeType, EdgeType definitions
    builder.ts       # Builds graph from a root path
    staleness.ts     # Mtime-based cache invalidation
    manager.ts       # Multi-environment graph manager
    overlay.ts       # Overlay scanning and graph merge
  tools/
    stats.ts         # graph_stats handler
    consumers.ts     # graph_find_consumers handler
    describe.ts      # graph_describe_pipeline handler
    impact.ts        # graph_impact_analysis handler
    lineage.ts       # graph_data_lineage handler
    paths.ts         # graph_find_paths handler
    search.ts        # graph_search_queries handler
    diff.ts          # graph_diff_pipeline handler
  parsers/
    pipeline.ts      # ADF pipeline JSON parser
    dataset.ts       # ADF dataset JSON parser
    columns.ts       # Column mapping extractor
    sql.ts           # SQL stored procedure parser
tests/
  graph/             # Graph model unit tests
  parsers/           # Parser unit tests
  tools/             # Tool handler unit tests
  integration.test.ts  # End-to-end test with real pipeline data

Dev Commands

npm run build        # Compile TypeScript
npm run dev          # Watch mode
npm test             # Run all tests (vitest)
npm run test:watch   # Watch mode tests