This guide is for developers and AI agents who want to set up SpecLoom quickly to enforce the Spec-Driven Development (SDD) process using the Hybrid Workflow paradigm.
- Node.js v20+
- Git
Run the following command to install the CLI tool globally:
npm install -g specloomInitialize the repository with a command.
loom initThis creates the .spec/ directory structure containing the system requirements, foundational context, and core protocols.
SpecLoom works best when an AI agent handles the heavy lifting via MCP Tools while humans make strategic decisions, reviews at Anchor Points.
- Initiation: Human runs
/loadto load the current state of the environment into the chat. - Product Definition: Product Owner uses
/visionand Analyst uses/req. AI drafts the requiredURandFRartifacts. - Handshake 1: Human runs
/handshaketo review the drafts and agree on the Product Anchor. - Architecture: Developer runs
/arch. AI drafts logical/physical components, functional chains, and API contracts based on the approved requirements. - Handshake 2: Human runs
/handshaketo agree on the Architecture Anchor. - Execution: PO reprioritizes the backlog using
/prioritize. Developers run/implto get the next coding task.
Notice that you do not need to run MCP commands (like /vision or /arch) manually; the AI agent will get all needed protocols from the task context.
To use SpecLoom as a "Brain" for your AI agent (Cursor, Windsurf, Cline), you must attach the MCP server to your local workspace.
Add the following to your agent's configuration file:
{
"mcpServers": {
"specloom": {
"command": "npx",
"args": [
"-y",
"--package",
"specloom",
"loom-server"
]
}
}
}- Orphans? Run
loom validateto find missing links. - Locked? Run
loom statusto see who holds the lock. - What's Next? If stuck, run
/loadvia MCP to have the AI suggest the next immediate action based on the state machine.