NexTool MCP for GLPI

Open-source Model Context Protocol server that connects AI assistants (Claude, ChatGPT, Cursor, VS Code, Windsurf…) to your GLPI service desk: tickets, ITIL objects (problems, changes), assets, knowledge base, users, rules and webhooks. It works with GLPI 10 and 11, and supports both the REST API v1 and the GLPI 11 API v2 in one server.

Not affiliated with Teclib'. GLPI is a registered trademark of Teclib'.

Two ways to use it:

• Hosted, early access: nothing to install. Sign in with a NexTool account, then add one URL as a connector in Claude or ChatGPT. Request access at https://nextoolsolutions.com/mcp. How it works, step by step: https://nextoolsolutions.com/en/mcp/como-funciona.
• Local, open source (this repository): runs on your machine with your GLPI credentials. It is free and also works with GLPI instances that are only reachable on your intranet.

1. Enable the REST API: Setup → General → API. Turn on "Enable REST API" and "Enable login with external token".
2. API client (optional but recommended): in the same screen, add an API client and copy its App-Token. If you restrict it by IP, allow the address the MCP connects from.
3. User token: Administration → Users → (the user) → Remote access keys. Copy the API token. Every action runs as this user and follows their GLPI profile and entities. Use a dedicated user with only the rights the assistant needs.
4. GLPI 11 API v2 (optional): create an OAuth client in GLPI 11 and note its client ID and secret. You also need a username and password, for the password grant.

Published on npm as @nextoolsolutions/mcp-glpi and listed in the official MCP Registry as com.nextoolsolutions/glpi. Requires Node.js 18+.

GLPI_URL=https://glpi.example.com GLPI_USER_TOKEN=xxx GLPI_APP_TOKEN=yyy npx -y @nextoolsolutions/mcp-glpi

From source instead: clone this repository, then npm ci && npm run build in mcp-glpi-core and in mcp-glpi, and run node mcp-glpi/dist/index.js.

All options are listed in mcp-glpi/README.md.

claude mcp add glpi \
  -e GLPI_URL=https://glpi.example.com -e GLPI_USER_TOKEN=xxx -e GLPI_APP_TOKEN=yyy -e GLPI_TOOLSETS=core \
  -- npx -y @nextoolsolutions/mcp-glpi

Edit claude_desktop_config.json (Settings → Developer → Edit config) and restart Claude Desktop:

{
  "mcpServers": {
    "glpi": {
      "command": "npx",
      "args": ["-y", "@nextoolsolutions/mcp-glpi"],
      "env": {
        "GLPI_URL": "https://glpi.example.com",
        "GLPI_USER_TOKEN": "xxx",
        "GLPI_APP_TOKEN": "yyy",
        "GLPI_TOOLSETS": "core"
      }
    }
  }
}

Use the same command/args/env block:

• Cursor: ~/.cursor/mcp.json (or .cursor/mcp.json in a project).
• Windsurf: ~/.codeium/windsurf/mcp_config.json.
• VS Code: .vscode/mcp.json. The top-level key there is "servers" instead of "mcpServers", and each entry also takes "type": "stdio".

The Dockerfile builds the stdio server from this repository (node:22-alpine, no credentials inside). Without GLPI_* variables it still starts and lists its tools, which is what directories that inspect servers need.

docker build -t nextool-mcp-glpi .
docker run -i --rm -e GLPI_URL=https://glpi.example.com -e GLPI_USER_TOKEN=xxx -e GLPI_TOOLSETS=core nextool-mcp-glpi

To serve several users or GLPI instances from one machine, run mcp-glpi/dist/http.js (Streamable HTTP, Bearer keys stored as SHA-256 hashes). See "Two ways to run" in the package README.

The hosted NexTool MCP lives at https://mcp.nextoolsolutions.com/mcp. It adds:

• sign-in through the NexTool portal (OAuth 2.1);
• one-click connectors for Claude and ChatGPT;
• GLPI credentials encrypted so that only the connector can use them;
• per-user rate limits and audit logs, without storing prompts or results.

Your GLPI must be reachable from the internet. If it only accepts known IPs, the portal shows the address to allow when you test the connection.

Request early access: https://nextoolsolutions.com/mcp. Once you are in:

Client	How to connect
Claude (web, desktop, mobile)	Settings → Connectors → Add custom connector → URL https://mcp.nextoolsolutions.com/mcp → sign in with your NexTool account and authorize
ChatGPT	Add a connector/app with the same URL (developer mode while the app is in review)
Claude Code	claude mcp add --transport http nextool https://mcp.nextoolsolutions.com/mcp, then /mcp to sign in. Or use an API key from My account → NexTool MCP: add --header "Authorization: Bearer nxm_…"
Cursor	(server URL https://mcp.nextoolsolutions.com/mcp, sign-in by OAuth), or the same URL with header Authorization: Bearer nxm_…
VS Code	Server URL https://mcp.nextoolsolutions.com/mcp with header Authorization: Bearer nxm_…

Connect your GLPI once at app.nextoolsolutions.com → My account → NexTool MCP (GLPI URL plus the tokens from step 1). Then click Test connection.

• Every tool carries MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint), so clients can ask before changing data.
• Deletes are refused unless GLPI_ALLOW_DELETE=true (always refused on the hosted free plan).
• GLPI_READ_ONLY=true blocks every write.
• Identical create calls within a short window are replayed, not repeated.
• Redirects are never followed. The API v2 client talks only to the configured GLPI origin.

• MCP Registry: com.nextoolsolutions/glpi (mcp-glpi/server.json): the npm package (stdio) and the hosted server as a remote.
• Cursor plugin: .cursor-plugin/plugin.json + mcp.json (hosted server, OAuth). Direct install link: cursor://anysphere.cursor-deeplink/mcp/install?name=nextool-glpi&config=eyJ1cmwiOiJodHRwczovL21jcC5uZXh0b29sc29sdXRpb25zLmNvbS9tY3AifQ==.
• Glama: glama.json names the maintainers; the Dockerfile is the build they run.

Pushing a tag v<mcp-glpi version> runs release.yml: tests, npm (versions already published are skipped), then the MCP Registry. Repository secrets:

cd mcp-glpi-core && npm ci && npm run build && npm test
cd ../mcp-glpi && npm ci && npm run build && npm test

This repository is a mirror published from NexTool's internal monorepo. Issues and pull requests are welcome here. Security reports: see SECURITY.md.

MIT © NexTool Solutions