Stigg MCP — Connect Once, Use Everywhere

The Stigg MCP server at https://mcp.stigg.io exposes a Stigg environment to any MCP-compatible AI client. The agent can then provision customers, manage subscriptions, check entitlements, and report usage — in natural language — without writing API calls.

Status: the Stigg MCP server is currently in public beta. Re-confirm via docs.stigg.io before relying on edge behavior.

Before You Start

Per the umbrella stigg skill: search first. If the user already has the Mintlify Stigg docs MCP available (search_stigg), use it to confirm the latest install command. Re-fetch the MCP-server docs page when in doubt.

The 3-Step Setup

Step 0 — You need a Stigg account

If you don't already have one: sign up at app.stigg.io. New accounts come with a default sandbox environment. Production environments are on the Scale plan.

If your agent runs the install command before Step 0 is done, the smoke test will fail with 401 Unauthenticated.

Step 1 — Get an API key

1. In the Stigg app, go to Integrations → API keys.
2. Pick the environment (sandbox / staging / production). Default to sandbox for AI-assisted dev.
3. Either copy the Full access key or click + Add scoped key (Scale plan) and grant only the permissions the agent needs. Treat the key like a password.

For better isolation, create a dedicated scoped key per AI client (e.g., "Claude Code – Dev", "Cursor – Staging").

Step 2 — Configure the client (snippets per AI client)

Claude Code (HTTP transport — preferred)

claude mcp add stigg \
  --header "X-API-KEY: <YOUR_API_KEY>" \
  --transport http https://mcp.stigg.io

Then start a new Claude Code session and run /mcp to confirm stigg is connected and its tools are listed.

Claude Desktop

Edit:

• macOS: ~/Library/Application Support/Claude/claudedesktopconfig.json
• Windows: %APPDATA%\Claude\claudedesktopconfig.json

Add under mcpServers:

{
  "mcpServers": {
    "stigg": {
      "url": "https://mcp.stigg.io",
      "headers": {
        "X-API-KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Restart Claude Desktop.

ChatGPT

Settings → Connectors, add a new MCP server:

Cursor / VS Code / Windsurf / Codex / generic clients (stdio bridge)

These clients use the local stdio bridge @stigg/typescript-mcp (run via npx). Edit the relevant config file (~/.cursor/mcp.json for Cursor; MCP: Open User Configuration in VS Code; the equivalent for Windsurf/Codex), and add:

{
  "mcpServers": {
    "stigg_typescript_api": {
      "command": "npx",
      "args": ["-y", "@stigg/typescript-mcp"],
      "env": {
        "STIGG_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Restart the client. For any client that supports HTTP transport natively, prefer the mcp.stigg.io HTTP form (same shape as the Claude Desktop snippet).

The canonical, always-current install matrix lives at docs.stigg.io/api-and-sdks/mcp-server. Refresh from there if a snippet here doesn't work.

Smoke Test

After install, confirm:

1. Tool listing. In Claude Code: /mcp — stigg should appear with its tools.
2. Search docs. Ask: "Search Stigg docs for 'create a customer'." If the agent invokes search_docs (or equivalent) and returns a doc snippet, transport is alive.
3. No-op read. Ask: "List the first 3 customers in this Stigg environment." This exercises a read against the live env. Expected on a fresh sandbox: an empty list — that's a pass.

If any step fails, the most common causes are: wrong header name (X-API-KEY, not Authorization), key from a different environment than expected, or the client wasn't restarted after editing the config.

When NOT to Use This Skill

Don't load this skill — go straight to a different one — when:

• You're writing production hot paths (gating, per-request usage reports). The MCP is non-deterministic by design; use the Node SDK instead — see stigg-api.
• The user explicitly asks for the CLI, or the task is unambiguously a deterministic shell workflow (CI, scheduled cron, runbooks). Use the Stigg CLI (brew install stiggio/tools/stigg — https://github.com/stiggio/stigg-cli) — see references/cli-vs-mcp.md. Otherwise default to the MCP server even for one-off ops.
• The user's question is about the REST or GraphQL API itself, not the MCP server — see stigg-api.

The MCP server fits when you're inside an AI assistant doing modeling, exploration, or one-off ops. It doesn't fit unattended automation or production hot paths.

Security — Read This Before Connecting Production

• Default to sandbox. AI agents can take destructive, hard-to-reverse actions based on natural-language instructions that may be misinterpreted. Develop and test against sandbox first.
• One scoped key per AI client. Don't reuse the full-access key everywhere. If a key leaks, blast radius = its scoped permissions only.
• Keys are environment-bound. A sandbox key cannot touch production data, and vice versa.
• Never commit keys. Treat them like passwords.
• Rotate periodically from Settings → API keys. Stigg supports zero-downtime rotation with a configurable grace period.
• The agent acts on your behalf. There is no separate agent identity. The MCP server forwards API calls using your key, scoped to its permissions.

Multi-Environment

Each mcpServers entry is per-environment. To switch, configure separate entries (stigg-sandbox, stigg-production) with separate keys. This is error-prone — the agent may not know which to target. Prefer running with a single environment per session and only adding the key for the environment you intend the agent to operate on.

What the Agent Can Do Once Connected

A non-exhaustive map (full surface continues to grow — re-check the docs page):

• Catalog: create / update / archive / list plans, features, add-ons; publish and archive packages.
• Customers & subscriptions: provision customers, provision/preview/cancel subscriptions (including scheduled and trial), grant promotional entitlements.
• Entitlements & usage: check entitlements, report usage measurements, report metering events.

A current, terse tool listing is in references/mcp-tools.md. Do not memorize this list — re-fetch from the live docs when in doubt.

Common Mistakes