Any MCP client works here.

Before you begin

You'll need a Stable Baseline API key. Visit the MCP setup guide to create one from your workspace settings. Keys are prefixed sta_… and shown only once.

Step 1 · Server details

Every client needs the same four pieces of information. Whichever config shape yours uses, the underlying values are identical:

Step 2 · Pick the format your client expects

Configuration formats in MCP land have diverged a little. Pick whichever matches your client — or use the stdio bridge at the bottom if yours doesn't speak HTTP at all.

Format A · JSON with mcpServers

Used by Cursor, Warp, Antigravity, and Claude Desktop.

{
  "mcpServers": {
    "sb": {
      "url": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Format B · JSON with servers

Used by VS Code with GitHub Copilot.

{
  "servers": {
    "sb": {
      "type": "http",
      "url": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Format C · JSON with serverUrl

Used by Windsurf. Note the serverUrl key — this is the single difference from Format A.

{
  "mcpServers": {
    "sb": {
      "serverUrl": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Format D · TOML

Used by OpenAI Codex.

[mcp_servers.sb]
url = "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp"
bearer_token_env_var = "STABLE_BASELINE_API_KEY"
http_headers = {}
enabled = true

Format E · Claude Code CLI

claude mcp add sb \
  --transport http \
  --url "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp" \
  --header "Authorization: Bearer YOUR_API_KEY"

Format F · OpenCode YAML

mcp:
  sb:
    type: remote
    url: https://api.stablebaseline.io/functions/v1/cloud-serve/mcp
    headers:
      Authorization: Bearer YOUR_API_KEY

Format G · stdio bridge fallback

If your client only supports stdio MCP servers, use the mcp-remote npm package as a bridge:

{
  "mcpServers": {
    "sb": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
        "--header",
        "Authorization: Bearer YOUR_API_KEY"
      ]
    }
  }
}

Step 3 · Generate your documentation

Once the MCP server is connected, open your client's AI chat or agent and type:

Run the sb-setup prompt from the sb server

The agent walks you through an interactive setup process to generate documentation for your project.

What sb-setup does — the 9-step process

• Verifies MCP connection — Confirms the Stable Baseline server is reachable and authenticated
• Resolves workspace & project — Finds or creates your Stable Baseline workspace and project
• Scans your codebase — Analyses project structure, tech stack, dependencies, and architecture
• Checks for existing docs — Detects if documentation already exists in Stable Baseline and offers to augment, replace, or cancel
• Creates folder structure — Organises documentation into logical folders matching your codebase
• Generates documents — Writes comprehensive docs for each area of your codebase
• Adds diagrams — Creates architecture and flow diagrams using Mermaid, PlantUML, or other supported formats
• Configures AGENTS.md — Adds auto-sync rules to your project's AGENTS.md file (augments existing content — never removes what's already there)
• Summary & next steps — Provides a summary of everything created with links to view in Stable Baseline

What gets documented

• Architecture — High-level system design, component relationships, data flow
• API & Integrations — Endpoints, external services, authentication flows
• Data Model — Database schemas, relationships, migrations
• Frontend — Component hierarchy, state management, routing
• Backend — Server logic, middleware, business rules
• DevOps — Build pipeline, deployment, environment configuration
• Auth & Identity — Authentication, authorisation, user management

The exact structure is determined dynamically based on your codebase — no two projects get the same output.

How auto-sync works

After setup, your AGENTS.md file contains rules that tell your AI agent when to update documentation:

Whenever your AI agent modifies files matching these patterns, it automatically updates the corresponding Stable Baseline documentation via the sb-sync prompt.

File structure

Your Stable Baseline Project
├── Architecture/
│   ├── High Level Overview
│   └── System Architecture Diagram
├── Frontend/
│   ├── Component Hierarchy
│   ├── State Management
│   └── Routing & Navigation
├── Backend/
│   ├── API Endpoints
│   ├── Business Logic
│   └── Middleware & Auth
├── Data Model/
│   ├── Database Schema
│   └── Entity Relationships
└── DevOps/
    ├── Build & Deploy
    └── Environment Config

This is an example. Your actual structure will be tailored to your specific project.

Keep docs updated mid-conversation

After making significant changes during a coding session — architectural decisions, new features, schema changes, key refactors — your agent will typically automatically propose updates to your documentation via the sb-update prompt based on the rules in your AGENTS.md file. Wait for the agent to finish its current coding step and it will often follow up by running sb-update on its own.

You can also manually capture everything in your documentation without leaving the conversation:

Run the sb-update prompt

The sb-update prompt scans the current conversation context, identifies what changed, maps changes to existing documents, and applies targeted patches.

Try it

Once setup is complete, try these prompts in your agent:

Regenerate documentation

To regenerate your entire documentation from scratch:

Run the sb-setup prompt from the sb server

Setup will detect existing documentation and ask whether you want to augment (add to existing), replace (start fresh), or cancel.

MCP compatibility

Stable Baseline's MCP server uses the Streamable HTTP transport and supports:

Other prompts available