Connect your agent over MCP

Chat apps in under a minute

claude.ai, Claude Desktop and other chat apps with custom connectors sign in with your Staglight account (OAuth 2.1), so nobody pastes a key. You approve each app once and give it a spending cap.

Claude

Add to Claude

1. Press the button above, or in Claude open Customize > Connectors > Add custom connector with the name Staglight and the URL https://staglight.ai/mcp.
2. Press Continue, keep Sign in now and press Add, then sign in with your Staglight account, choose what it may do and its spending cap, then Allow. On the Free plan you can add one custom connector; on Team and Enterprise an owner adds it first and you press Connect.
3. In a chat, press + > Connectors and switch Staglight on. Attach a product photo and ask: make me an ad for this product.

ChatGPT

1. In ChatGPT on the web, open Settings > Plugins and press + > Add custom MCP server. It needs a Business, Enterprise or Edu workspace that allows it; on Pro it can only read.
2. Name it Staglight, paste the URL https://staglight.ai/mcp, choose OAuth, tick I understand and want to continue, then Create as a plugin.
3. Sign in with your Staglight account and Allow.
4. In a new chat, type @, pick Staglight and ask for your ad.

Every connection is listed in Settings > Connections with its spending cap (any 24 hours); Disconnect ends its access at once. Tools that spend, upload or change things need the Create ads and spend credits permission; a view-only connection can read but not spend.

Server

Create a key with a daily cap in the studio and save it as STAGLIGHT_API_KEY.

export STAGLIGHT_API_KEY="sl_live_..."

Claude Code

One command, for every project on this computer. The shell fills in the key and Claude Code stores it in ~/.claude.json.

claude mcp add --transport http staglight https://staglight.ai/mcp \
  --header "Authorization: Bearer $STAGLIGHT_API_KEY" --scope user

Prefer to sign in instead of using a key? Run the command without the --header part, then type /mcp in Claude Code and pick staglight to sign in with your Staglight account.

Check that it connected:

claude mcp get staglight

To share it with a team in the repo, add .mcp.json. Each person keeps the key in their own STAGLIGHT_API_KEY.

{
  "mcpServers": {
    "staglight": {
      "type": "http",
      "url": "https://staglight.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${STAGLIGHT_API_KEY}"
      },
      "timeout": 600000
    }
  }
}

Name the variable STAGLIGHT_API_KEY. Claude Code leaves some credential variables with reserved names empty in remote headers, so use this one.

Codex

codex mcp add staglight --url https://staglight.ai/mcp --bearer-token-env-var STAGLIGHT_API_KEY

Or write it by hand in the Codex config:

[mcp_servers.staglight]
url = "https://staglight.ai/mcp"
bearer_token_env_var = "STAGLIGHT_API_KEY"
startup_timeout_sec = 20
tool_timeout_sec = 120

Codex stops a tool call after 60 s by default. Every tool answers in under 50 s, and get_job waits 50 s at most.

Cursor

In ~/.cursor/mcp.json for every project, or .cursor/mcp.json for one project:

{
  "mcpServers": {
    "staglight": {
      "url": "https://staglight.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${env:STAGLIGHT_API_KEY}"
      }
    }
  }
}

Claude Desktop and claude.ai

claude.ai and Claude Desktop connect by signing in with your Staglight account: see Chat apps.

If you'd rather use an API key there, bridge from your computer with the community package mcp-remote in claude_desktop_config.json:

{
  "mcpServers": {
    "staglight": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://staglight.ai/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}",
        "--transport",
        "http-only"
      ],
      "env": {
        "AUTH_HEADER": "Bearer sl_live_..."
      }
    }
  }
}

mcp-remote is a third-party package that runs on your computer with your key. Review it before you use it.

Tools

Paid tools take max_credits and idempotency_key. Every result has a short text, the data as JSON and, for finished files, links to download.

Money rules for agents

• Call estimate_cost before any paid tool. It's free and can price the exact call: tool plus arguments.
• Send max_credits and a fresh idempotency_key with every paid call.
• Paid tools answer at once with a job_id or run_id. Poll get_job or get_ad with wait_seconds=50.
• After a timeout, call list_jobs with the same idempotency_key before paying again.
• Failed jobs are refunded automatically. Out of credits, buy_credits_link gives the user a link to the studio. No payments go through MCP.
• Give every agent its own key with a daily cap. At the cap it gets daily_limit_reached and stops.
• With auto: true, create_ad and create_ad_from_brief approve each review step on their own while the whole run stays within max_credits. If the next step would go over, the run waits at that step with auto_stopped and nothing more is charged.

Errors

Tool errors come back as a result with isError: true and the same codes as the REST API. Out of credits includes needed, balance and topup_url. A missing or wrong key is HTTP 401 with WWW-Authenticate.

{
  "content": [
    {
      "type": "text",
      "text": "Error insufficient_credits: This needs 63 credits and the balance is 10. ..."
    }
  ],
  "structuredContent": {
    "error": {
      "code": "insufficient_credits",
      "message": "This needs 63 credits and the balance is 10. Buy credits to continue.",
      "status": 402,
      "needed": 63,
      "balance": 10,
      "charged": false,
      "topup_url": "https://staglight.ai/app/creditos?need=53"
    }
  },
  "isError": true
}

Test with curl

The server answers plain JSON over POST, with no sessions, so curl works:

curl -s https://staglight.ai/mcp \
  -H "Authorization: Bearer $STAGLIGHT_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {"protocolVersion": "2025-11-25", "capabilities": {}, "clientInfo": {"name": "curl", "version": "1.0"}}}'

curl -s https://staglight.ai/mcp \
  -H "Authorization: Bearer $STAGLIGHT_API_KEY" \
  -H "Content-Type: application/json" \
  -H "MCP-Protocol-Version: 2025-11-25" \
  -d '{"jsonrpc": "2.0", "id": 2, "method": "tools/call", "params": {"name": "get_balance", "arguments": {}}}'

This page as Markdown · llms.txt