fluent-graph-mcp

MCP server for fluent-graph. lets your AI tools query your ServiceNow artifact dependency graph before making code changes.

Connect Claude Desktop, Cursor, or any MCP-compatible agent to your project's graph via a single config line. The agent calls blast_radius before touching any artifact and knows exactly what else breaks — with zero procurement friction.

Prerequisites

• Node.js 18+
• A ServiceNow Fluent SDK project

@yesprasad/fluent-graph does not need to be installed in the project. The MCP server resolves it in this order:

1. <project>/node_modules/.bin/fluent-graph — used if present (fastest)
2. npx @yesprasad/fluent-graph — used otherwise; finds a global install or downloads it on first run

The simplest setup is a one-time global install:

npm install -g @yesprasad/fluent-graph

Install

npm install -g fluent-graph-mcp

Configure

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "fluent-graph": {
      "command": "fluent-graph-mcp",
      "args": ["--project-root", "/path/to/my-servicenow-project"]
    }
  }
}

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "fluent-graph": {
      "command": "fluent-graph-mcp",
      "args": ["--project-root", "/path/to/my-servicenow-project"]
    }
  }
}

How it works

On startup, fluent-graph-mcp runs fluent-graph analyze in your project root, loads the resulting fluent-graph.json into memory, and starts an MCP stdio server. All tool queries are served from in-memory (fast). Re-extraction only happens when you call refresh.

Tools

blast_radius

Returns all ServiceNow artifacts impacted if the named artifact changes, grouped by hop distance.

| Arg | Type | Required | Description | |---|---|---|---| | artifact_name | string | yes | The artifact table name to check | | depth | number | no | Max hop depth (default 5, max 25) |

Example output:

Blast radius for: Incident Severity Alert Flow
Total impacted: 7 artifacts across 2 hops

Hop 1 (direct dependents):
  - Laptop Request (table: sc_cat_item) via: flow_designer_flow

Hop 2 (transitive):
  - expedited (table: item_option_new) via: cat_item
  - preferredModel (table: item_option_new) via: cat_item

get_artifact

Returns details of a named artifact: its table, action, source file, attributes, and all outgoing dependency edges.

| Arg | Type | Required | Description | |---|---|---|---| | name | string | yes | Artifact name to look up |

list_artifacts

Lists all artifact names in the project. Optionally filter by table name.

| Arg | Type | Required | Description | |---|---|---|---| | table | string | no | Filter by ServiceNow table (e.g. sys_script_include) |

refresh

Re-runs fluent-graph analyze and reloads the in-memory graph. Call this after making code changes.

Returns: Graph refreshed. 215 nodes, 193 edges loaded.

The agent workflow

1. Agent calls list_artifacts → discovers what's in the project
2. Agent calls blast_radius before touching anything → knows what breaks
3. Agent makes changes to code
4. Agent calls refresh → graph is up to date
5. Agent calls blast_radius again → validates change is safe