The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external systems — data sources, tools, and workflows. Anthropic introduced it in late 2024, and it is now supported by clients such as Claude and ChatGPT, as well as developer tools like VS Code and Cursor. Think of it as USB-C for AI: instead of a bespoke connector for every integration, there is one standardized way for an assistant to see into your system and act on it.
For service management that has a concrete consequence. GLPI has had a REST API for years, and an MCP server lets you talk to your tickets in plain language — straight from a conversation with the assistant.
What connecting to GLPI enables today
Community MCP servers for GLPI (for example the svtica/glpi-mcp or GMS64260/mcp-glpi projects) expose everyday service desk operations to the assistant:
- Creating, reading, and updating tickets, including linking them.
- Adding followups, tasks, and solutions that close a ticket.
- Working with the knowledge base, including article visibility controls.
- Statistics — tickets by status, type, priority, or SLA compliance.
In practice, instead of clicking through forms you tell the assistant "summarize this week's open incidents that breached SLA" or "open a ticket and assign it to the operations team," and it does so through the API. Routine but time-consuming steps shrink from minutes to seconds.
How it works technically
The MCP server sits between the assistant and GLPI and translates requests into REST API calls. GLPI 10 uses apirest.php; GLPI 11 keeps that interface and also exposes it under the api.php/v1 path. Authentication runs through an App-Token and a User-Token, from which GLPI issues a temporary Session-Token. The REST API must be enabled and an API client created in GLPI settings. GLPI 11's new high-level API (api.php/v2) is a separate interface authenticated via OAuth2 — today's MCP servers still build on the older one.
Official versus community route
It helps to separate two worlds. GLPI also offers an official GLPI-AI plugin (available with a GLPI Network subscription) that can summarize a ticket's timeline via OpenAI. That is a narrowly scoped feature straight from the maintainers. The MCP servers above, by contrast, are unofficial community projects — they give a far broader range of operations, but responsibility for security and maintenance sits with you.
What to watch
- Least privilege — scope the API account the assistant acts through with a GLPI profile and RBAC so it can only touch what it genuinely needs. Do not run it as an administrator.
- On-premises operation — with a self-hosted GLPI and MCP server, your data leaves your environment no more than it has to.
- Audit and logging — every action the assistant takes remains an entry in the ticket history; for regulated entities under NIS2 that is worth tracking.
- Code review — vet an unofficial server before deployment, as you would any integration with access to production.
MCP is not a demo toy — it is a standard fast becoming a common layer between AI and enterprise systems. With GLPI, a sensible start is simple: a tightly scoped account, reading and summaries first, and only after validation a gradual move to actions that change data.