Developer Setup Guide

← Back to Documentation

Before you start: Your enterprise admin should have created a grant for you and shared an invite secret (a one-time code, used the very first time you log in — you'll set a password after that). You'll also need the enterprise slug, your member ID, and the project ID for the project you're working on. Check your email — these may have been sent to you.

Install the Corla adapter

Install the adapter globally using npm:

npm install -g @corla/adapter

This gives you the corla CLI command. Verify the installation:

corla status

Log in (password or invite secret)

Run the login command:

corla login

You'll be prompted for the broker URL (https://broker.corla.ai), your enterprise slug, member ID, and project ID — your admin will have shared these. Then for the credential.

First login — invite secret + set a password:

Enter the one-time invite secret from your admin.
The CLI prompts you to set a password. This password is yours from then on.

Subsequent logins — password only:

corla login auto-detects your existing account and asks for your password — no admin involvement needed to reconnect.
Forgot or want to rotate it? corla set-password sets a new one from the terminal.

On success, you'll receive a token. The CLI prints the export command:

export CORLA_TOKEN=eyJhbGci...your-token-here

Add it to your shell profile (~/.zshrc or ~/.bashrc) so it persists across terminal sessions. Tokens expire; corla refresh silently rotates an expiring token without opening a browser or re-prompting.

Prefer a browser? Sign in at admin.corla.ai/member/login — the member portal. View your active projects, change your password, copy your token, and pull the connector instructions for Claude.ai or ChatGPT in one place.

Keep your token private. Never commit it to git, share it in chat, or include it in screenshots. It grants access to your enterprise's context. If compromised, change your password (corla set-password) and ask your admin to revoke the active sessions.

Configure your project

Navigate to your project directory and run:

cd /path/to/your/project
corla init

This writes the MCP config for every supported IDE in one shot:

.mcp.json — root config (Claude Code, Cursor, Windsurf)
.vscode/mcp.json — VS Code Copilot

The .mcp.json looks like this:

{
  "mcpServers": {
    "corla": {
      "type": "http",
      "url": "https://broker.corla.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${CORLA_TOKEN}"
      }
    }
  }
}

Using Antigravity? Pass the flag and the Antigravity config gets written too:

corla init --antigravity

Start using enterprise context

That's it. Your AI tools will now automatically load enterprise context when working in this project. The context is fetched from the broker, compiled and scoped to your role, and delivered to your AI tool.

Claude Code

Context loads automatically via MCP

Cursor

Reads .mcp.json for context

VS Code Copilot

Uses .vscode/mcp.json

Windsurf

Reads .mcp.json for context

Claude.ai (web/mobile)

Custom connector — paste broker URL

ChatGPT (web/mobile)

Custom connector — paste broker URL

To verify the connection is working:

corla status

You should see your enterprise name, project, role, and token expiry.

Switching between projects

If your admin has granted you access to multiple projects, you can switch between them inside a single connection — no disconnect, no re-auth, no admin ticket.

Claude Code, Cursor, Claude.ai, and ChatGPT discover available projects automatically via the broker's list_projects MCP tool. Just ask the agent:

"Which projects do I have access to?"
"Switch to the payments project."

The session-level project context updates immediately. The next compiled package is scoped to the new project.

—

What happens under the hood

When your AI tool starts a session, it connects to the Corla broker via MCP and:

Your token is validated and checked against the revocation list
Your grant, role, and project are verified
Relevant assets are compiled — redacted, scoped to your role, and signed
The compiled context is delivered to your AI tool in-memory (never written to disk)
Every access is logged in an immutable audit trail

You never see the raw assets. Your AI tool receives processed context that helps it write better, more aligned code — without exposing the enterprise's proprietary material.

Troubleshooting

"Token expired" or "401 Unauthorized"
Your token has expired. Run corla refresh for a silent rotation (no browser, no re-auth). If the underlying session is gone, corla login with your password gets you a fresh one.

"Forgot my password"
Run corla set-password to set a new one from the terminal, or sign in to admin.corla.ai/member/login and reset it from the member portal. (If you've never set one, you'll still have your one-time invite secret — ask your admin to re-issue it if lost.)

"Grant not found" or "Policy denied"
Your grant may have been revoked or expired. Contact your enterprise admin to check your grant status.

"Invalid invite secret"
Invite secrets are single-use. If you've already used it, you don't need it again. If you haven't and it's not working, ask your admin to revoke the grant and create a new one.

MCP server not showing up in your AI tool
Make sure CORLA_TOKEN is set in your environment and .mcp.json exists in your project root. Restart your AI tool after adding the config.

Need help?

If you run into any issues, reach out to your enterprise admin or contact us at info@noitechnologies.com.