Get started
Connect your first agent in a few minutes.
This guide walks you from zero to a working MCP connection. We'll use Claude Code for the example, but the same steps apply to Cursor, Windsurf, and anything else that speaks MCP.
1. Create a workspace
Sign in at app.stablebaseline.io and create your first workspace. Name it after your team or company — it's the top-level container that holds your projects, members, and billing.
2. Create a project
Inside the workspace, create a project for whatever you want agents to document or manage. For a codebase-onboarding scenario, create a project named after your repo (e.g. acme-api). The project is the scope that holds your docs, diagrams, plans, improvements, and scans — agents do their work inside a project.
- Codebase onboarding — one project per repo or service. Agents create architecture docs, sequence diagrams, and onboarding guides right next to the code.
- Compliance readiness — one project per framework (e.g. SOC 2, ISO 27001). Agents cross-reference docs and surface gaps as improvements.
- Product / initiative — one project per product line or quarter, with phases and tasks for humans and agents to share.
First-timer tip
sb-setup, which analyses your repo and scaffolds a complete documentation structure inside the project automatically.3. Generate an MCP key
Your agent authenticates to Stable Baseline with a key. You have two ways to issue one, depending on how broad you want the agent's access to be.
Option A · Project-scoped key (recommended)
Open the project, then Project settings → MCP keys and click Generate new key. The key can only read and write inside this one project — useful when you want an agent tightly scoped to a single repo or initiative. Copy the key; it's shown once.
Option B · Global key or OAuth from your profile
If you want one key that covers everything you yourself can see — every workspace and every project you're a member of — open the profile menu in the top-right and choose Account settings → MCP keys. From there you can either generate a personal key or connect via OAuth, which signs the agent in as you (no key to manage, revocable from the same screen).
Treat keys like passwords
4. Configure your MCP client
Each client has a slightly different config file. For Claude Code, edit ~/.claude/claude_desktop_config.json:
{
"mcpServers": {
"stable-baseline": {
"url": "https://api.stablebaseline.io/functions/v1/cloud-serve/mcp",
"headers": {
"Authorization": "Bearer ${SB_API_KEY}"
}
}
}
}Restart Claude. The stable-baseline server should now appear in the connected servers list with all 60+ tools available.
5. Run your first task
Ask your agent:
Run sb-setup on this repo and create a documentation structure in my Stable Baseline project. Use Mermaid diagrams for the architecture of each service in src/.
Within a few minutes you'll have a folder hierarchy, README-style docs for each service, and a set of architecture diagrams — all stored inside the project you created in step 2 and ready for the next agent (or teammate) to pick up.
What's next
- Give the agent a scoped role so it can only touch certain projects.
- Invite a teammate and watch their edits appear alongside the agent's.
- Set up a plan with phases and tasks, then assign tasks to the agent.