Connect limena to Claude

Step 1 — Generate an API key

limena uses long-lived workspace API keys for agent integrations. You'll need admin role in your limena workspace.

Sign into console.limena.app
Go to Settings → API keys
Click Generate new key, name it (e.g. Claude desktop), pick scopes (Read + Write recommended for a conversational agent), and hit Generate
Copy the plaintext lmn_… immediately — it's only shown once

Step 2 — Add limena to your agent

Pick your agent. The connection target is the same in all cases — limena exposes seven tools (list_clients, create_client, list_programs, create_program, run_audit, get_audit, get_findings) via the Model Context Protocol.

Recommended · One-click

Claude.ai (web)

Claude.ai → Customize → Personal plugins
Click + → Create plugin → Add marketplace
Paste: https://limena.app/plugins.json
Install the limena — accessibility audits plugin
When prompted, paste your lmn_… key

Now in any conversation: "Use limena to audit example.com."

Native config

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "limena": {
      "transport": {
        "type": "http",
        "url": "https://console.limena.app/api/mcp"
      },
      "headers": {
        "Authorization": "Bearer lmn_..."
      }
    }
  }
}

Quit and reopen Claude Desktop.

Terminal

Claude Code

Export the key in your shell:

echo 'export LIMENA_API_KEY="lmn_..."' \
  >> ~/.zshenv
source ~/.zshenv

Install the limena skill (one-time):

mkdir -p ~/.claude/skills/limena
curl -fsSL https://limena.app/skill.md \
  -o ~/.claude/skills/limena/SKILL.md

Quit + relaunch Claude Code.

MCP-compatible

Cursor / Zed

Add to your editor's MCP config:

{
  "mcpServers": {
    "limena": {
      "transport": {
        "type": "http",
        "url": "https://console.limena.app/api/mcp"
      },
      "headers": {
        "Authorization": "Bearer lmn_..."
      }
    }
  }
}

Reload the editor's MCP servers.

Step 3 — Try it

In any conversation with your now-connected agent, try:

"Use limena to audit https://example.com."
"List my limena clients."
"Run a limena audit on all the pages in my pricing program."
"Show me the findings from my last limena audit."

limena's plugin walks you through "which client?" → "which program?" → "scan options?" → queue, poll, summarize. Same conversational flow on every agent.

Security and scopes

Every key carries its own scopes, optional origin allowlist, configurable rate limit (default 100/min), and a 30-day usage log. Same posture as Stripe, GitHub, or Linear API keys.

Read scope — list clients, programs, and read audit findings (safest for browser-based connectors)
Write scope — create clients, programs, queue audits
Admin scope — delete resources (reserved for cleanup tools; don't grant to conversational agents)

For Claude.ai web you can lock a key to https://claude.ai using the optional origin allowlist. For server-to-server CI tools, leave origins blank.

Revoke any key in Settings → API keys — calls stop in under a second.

Let Claude write into your accessibility memory.