Home / Docs / GAT – Grid Analysis Toolkit / Developer Internals / MCP Server Setup
2 min read

MCP onboarding for GAT

The MCP server should serve a ready-to-use tree the moment the machine is provisioned. The quick path is:

  1. Install Rust + workspace dependencies (see README.md).
  2. Run scripts/mcp-onboard.sh; it regenerates every doc target (cargo xtask --features docs doc all) and starts gat-mcp-docs --docs docs --addr 127.0.0.1:4321 so agents can immediately discover the latest CLI refs, schemas, and guide content.
  3. Visit http://127.0.0.1:4321/ (via MCP tooling) and open docs/mcp/manifest.json, which curates the most helpful commands, datasets, and guides. Use that manifest as your first checkpoint when learning or scripting with GAT.

Manifest highlights (docs/mcp/manifest.json)

The manifest is intentionally small, with these sections:

  • onboarding — the helper script that keeps the MCP server aligned with the repo.
  • commands — sample CLI invocations for DC/AC PF, analytics, DER prep, and reproducible runs.
  • datasets — curated dataset fetch commands that any MCP agent can execute before launching workflows.
  • guides — quick links into the ADMS, DERMS, and distribution docs that explain how gat-adms, gat-derms, and gat-dist tie into the CLI.

Explore that manifest and copy the commands you need; they will still work from any shell that has gat installed and points to this workspace.

Keeping everything fresh

After you make documentation or manifest changes:

  1. Regenerate the docs (cargo xtask --features docs doc all).
  2. Start (or restart) gat-mcp-docs --docs docs --addr 127.0.0.1:4321 so the new tree is served.
  3. The manifest is part of the docs tree, so any restart of scripts/mcp-onboard.sh keeps the commands and guides current.

Parallel workflow inspiration

These commands can run in parallel/automation contexts:

  • gat pf dc + gat pf ac to explore solver outputs per grid.
  • gat analytics ptdf to seed gat-dist or gat-adms heuristics.
  • gat dataset public fetch tasks that download telemetry/datasets before analytic runs.

Use parallel or xargs -P with the manifest entries to fan-out workloads right from the MCP tree.

Running manifest commands

Once MCP has started, the manifest entries become the canonical list of starter commands. Use scripts/mcp-manifest-run.sh commands to execute every command sequentially (or scripts/mcp-manifest-run.sh datasets to fetch the curated datasets). Pass an optional filter substring to run only the entries whose name matches (e.g., scripts/mcp-manifest-run.sh commands ptpf), add --dry-run to preview the commands, or pipe the manifest into parallel job runners if you want multiple hosts to share the same workload map.