Strapi Plugin AI Chat — AI skill for Claude Code

AI Chat for Strapi

AI chat inside the Strapi v5 admin, with a tool system your own plugins extend, and an MCP server that hands the same tools to external clients.

The chat is the surface people see. Underneath it is a tool registry: this plugin ships content tools out of the box, any other plugin can contribute its own through a small service contract, and everything registered is exposed to external AI clients through Strapi's built-in MCP server. One definition, three consumers.

Bring your own model. Anthropic, or anything OpenAI-compatible you run yourself.

---

Highlights

• Bring your own model. Anthropic, or open-source models you run yourself. Nothing assumes a vendor.
• Runs fully local. Qwen, Gemma, Llama and friends via Ollama, vLLM or llama-swap. Your content never leaves the machine. Read Choosing a model first, since model size decides whether tool calling actually works.
• Per-tool permissions. Grant searchContent without granting sendEmail, per role and per token.
• MCP built in. Registers its tools onto Strapi's official /mcp endpoint for Claude Desktop and other clients.
• Extensible. Other plugins contribute tools through a small service contract.
• Guardrails. Prompt-injection screening on chat input.
• Shows you the budget. A context badge reports how much of the model's window the conversation is using, which is the failure that otherwise looks like broken tool calling.
• Shows the thinking. Reasoning models stream their reasoning into a collapsible panel instead of pausing silently.

---

Overview

Strapi holds unpublished drafts, customer records and internal documents. This plugin gives a model tools that read and write them, which makes the choice of where inference happens a real decision rather than a detail. You can keep all of it inside your own network.

There are exactly two surfaces:

Surface	Where	Authenticates with	Tools it can use