SpecLoom

SpecLoom is an MCP Server and CLI that enforces the V-Model for AI-generated code, acting as a Guardian to prevent hallucinations, ensure traceability, and require that every line of code maps to a documented requirement. It exposes a set of MCP tools designed to work with AI agents to manage context, design, code, and verification across a project. When integrated with agents like Gemini CLI, Claude Desktop, Cursor, Windsurf, or Cline, SpecLoom provides structured Context Bundles (Requirements + Design + Code) and applies a four-eyes review process to maintain enterprise-grade compliance. The available MCP tools include loom_next (project planning prompts), loom_context (specs and code retrieval for a task), loom_validate (quality/consistency checks), and loom_verify (traceability and requirement conformance verification).

Prerequisites:

• Node.js and npm installed on your system

Step-by-step installation:

1. Install the Specloom CLI globally via npm:

npm install -g specloom

2. Confirm installation:

specloom --version

3. Start the MCP server (as configured in mcp_config):

# If using the provided MCP config, you can start via your MCP orchestrator which will invoke:
# command: npx, args: [-y, specloom, serve]

4. If you’re integrating with an agent, add the Specloom MCP server under the mcpServers configuration using the example below (see mcp_config).

Tips and considerations:

• Integration: SpecLoom is designed to work with various AI agents (Gemini CLI, Claude Desktop, Cursor, Windsurf, Cline). Use the loom_* tools to guide agents through context, design, and verification workflows.
• Four-Eyes: Ensure proper identity separation to prevent self-approval of code; configure roles and review processes in your workspace.
• Traceability: SpecLoom stores artifacts as JSON within your repository to maintain a Git-native workflow and audit trails.
• Deployment: If you need to run in containers or CI, adapt the mcp_config to your environment (e.g., docker or uvx/python runners) as needed.
• Troubleshooting: If MCP routing fails, verify that the host process has network access to the agent endpoints and that the specloom package is available in the npx registry.
• Documentation: Refer to the included docs (docs/ manual, architecture, methodology) for deeper integration guidance and best practices.