runfabric

RunFabric MCP Server

The RunFabric MCP (Model Context Protocol) server exposes plan, deploy, doctor, remove, and invoke as tools for AI agents and IDEs (e.g. Cursor, Claude Code).

What it does: Scope
Run it locally: Running the server
Connect from Cursor: Adding to Cursor
Tool shapes: Tool parameters

Scope

Tools: runfabric_doctor, runfabric_plan, runfabric_build, runfabric_deploy, runfabric_remove, runfabric_invoke, runfabric_logs, runfabric_list, runfabric_inspect, runfabric_releases, runfabric_generate, runfabric_state, runfabric_workflow
Transport: stdio (process-spawned). The client launches the server and communicates via stdin/stdout.
Requirement: The runfabric CLI must be on PATH. Build from repo with make build; the binary is at bin/runfabric. Override with RUNFABRIC_CMD if the binary has another name or path.

Running the server

From the repo root:

cd protocol/mcp && npm install && npm run build && node dist/index.js

Or after build, run the binary:

node protocol/mcp/dist/index.js

Ensure runfabric is on PATH (e.g. export PATH="$PWD/bin:$PATH" from repo root after make build).

Adding to Cursor

In Cursor settings (or .cursor/mcp.json), add an MCP server that runs the RunFabric MCP process:

Command: node
Args: path/to/unifn-framework-scaffold/protocol/mcp/dist/index.js
Env (optional): RUNFABRIC_CMD=/path/to/runfabric if the CLI is not on PATH

The exact configuration format depends on your Cursor/MCP client; use the stdio transport and point to the built dist/index.js (or npm start from protocol/mcp).

Tool parameters

Tool	Parameters
`runfabric_doctor`	`configPath` (optional), `stage` (optional)
`runfabric_plan`	`configPath`, `stage`, `provider` (optional)
`runfabric_build`	`configPath`, `stage`, `provider` (optional)
`runfabric_deploy`	`configPath`, `stage`, `preview`, `provider` (optional)
`runfabric_remove`	`configPath`, `stage`, `provider` (optional)
`runfabric_invoke`	`configPath`, `stage`, `function` (required), `payload` (optional), `provider` (optional)
`runfabric_logs`	`configPath`, `stage`, `function` (optional), `all` (optional), `provider` (optional)
`runfabric_list`	`configPath`, `stage`, `provider` (optional)
`runfabric_inspect`	`configPath`, `stage`, `provider` (optional)
`runfabric_releases`	`configPath`, `stage`, `provider` (optional)
`runfabric_generate`	`configPath` (optional), `stage` (optional), `args` (optional string array; forwarded after `runfabric generate`)
`runfabric_state`	`configPath` (optional), `stage` (optional), `command` (required), `args` (optional string array)
`runfabric_workflow`	`configPath` (optional), `stage` (optional), `command` (required), `args` (optional string array)

All tools run the CLI with --json, --non-interactive, and --yes so output is machine-readable and no prompts or confirmations appear. Results are the combined stdout/stderr as text content. When the exit code is non-zero, the result is marked with isError: true.