MCP Server Integration

The rrlmgraph-mcp companion package exposes the project knowledge graph via the Model Context Protocol (MCP), enabling any MCP-capable editor or AI agent to query graph context without running R code directly.

Architecture

┌─────────────────┐        JSON-RPC / stdio        ┌─────────────────┐
│   VS Code /     │ ◄────────────────────────────► │  rrlmgraph-mcp  │
│  GitHub Copilot │                                 │  (Node.js MCP   │
│   (MCP client)  │                                 │    server)      │
└─────────────────┘                                 └────────┬────────┘
                                                             │  SQLite
                                                    ┌────────▼────────┐
                                                    │  .rrlmgraph/    │
                                                    │  graph.sqlite   │
                                                    └─────────────────┘

The SQLite database is written by export_to_sqlite() in rrlmgraph and read at query time by the MCP server — no R process needs to be running during IDE sessions.

Installation

Install the MCP server globally via npm:

# In a terminal:
# npm install -g rrlmgraph-mcp

Or use npx for a one-off run:

# npx rrlmgraph-mcp --db-path path/to/.rrlmgraph/graph.sqlite

Exporting the graph to SQLite

library(rrlmgraph)

graph <- build_rrlm_graph("path/to/your/project")
export_to_sqlite(graph, "path/to/your/project/.rrlmgraph/graph.sqlite")

Configuring VS Code

Add the following to .vscode/mcp.json (or your user-level settings.json under github.copilot.chat.mcp.servers):

{
  "servers": {
    "rrlmgraph": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "rrlmgraph-mcp",
        "--db-path",
        "${workspaceFolder}/.rrlmgraph/graph.sqlite"
      ]
    }
  }
}

After saving, GitHub Copilot Chat will list rrlmgraph as an available MCP tool. Ask it questions like:

“Which functions call prepare_data()?” “What context is relevant for the data validation pipeline?”

Available MCP tools

Tool	Description
`query_context`	Relevance-guided BFS context window for a coding task
`get_node_info`	Full metadata, callers, callees, and source for a named node
`list_functions`	Functions ranked by PageRank, with optional file-path filter
`find_callers`	All functions that call a given function (incoming CALLS edges)
`find_callees`	All functions called by a given function (outgoing CALLS edges)
`search_nodes`	Full-text keyword search across node names, bodies, and docs
`add_task_trace`	Record accepted / rejected task outcome to improve future retrieval
`rebuild_graph`	Trigger an Rscript rebuild of the project graph in-process

Keeping the database fresh

Re-run export_to_sqlite() after editing R source files, or use update_graph_incremental() followed by export_to_sqlite() for large projects:

graph <- update_graph_incremental(
  graph,
  changed_files = "R/my_function.R"
)
export_to_sqlite(graph, "path/to/your/project/.rrlmgraph/graph.sqlite")

A targets pipeline or a Git pre-commit hook that runs this two-step command keeps the graph in sync with every commit.