Skip to main content
plamotrack ships with a Model Context Protocol (MCP) server built in — same process, no separate service to install or run. Point Claude, ChatGPT, Gemini Spark, Mistral or any MCP-compatible AI assistant at your instance and you can manage your collection in natural language: log an order from an email, mark a parcel received, check what’s on your bench — all by talking.
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external services. plamotrack implements it natively — no plugin, no middleware, no separate process.

​
What you can do

Your AI assistant has access to over 30 tools covering your entire collection. Here are a few examples of what that looks like in practice:
  • “The Sinanju arrived” → Claude calls mark_order_received, the kit moves to backlog, and your Gundam Markers are on hand.
  • “Add that Gundam Express order from my email” → Claude reads the email, calls create_order, and kits and supplies are logged against a new pending order.
  • “What’s on my bench?” → Claude calls list_kits filtered to Building status and tells you exactly what you’re supposed to be working on.

​
Two ways to connect

Personal access token

Create a token in Settings → Access tokens, then paste it into your AI client’s config. Works with Claude Desktop, Claude Code, and any client that can send an HTTP header. Best for local and LAN instances.

OAuth sign-in

For Claude.ai, ChatGPT, Gemini Spark and Mistral on the web. Requires OIDC mode and TLS. You sign in through your identity provider — no token to paste or store.

​
Which clients work

Every client below was connected to a plamotrack instance behind TLS on the current release, then used to list the tools and call one. For Claude Desktop, that check drove the mcp-remote bridge its guide configures.

​
Protocol versions

The MCP protocol has two generations in use: the current revision, where each request stands on its own, and the earlier one, where a client opens a session with a handshake first. plamotrack serves both from the same /mcp/ endpoint and lets each client choose, request by request. There is nothing to configure and nothing to keep in step: an older client keeps working when the assistant behind it updates, and a newer one gets the newer transport on its first call.
At the time of writing, Claude.ai and Claude Code use the current revision; ChatGPT, Gemini Spark, Mistral and MCP Inspector use the previous one. Either way, the tools and what they do are identical.

​
The MCP endpoint

The MCP endpoint is at <your instance>/mcp/. Always include the trailing slash.
If you reach plamotrack by a name other than localhost — a LAN hostname, a container name, a WireGuard mesh address — add that name to ALLOWED_HOSTS in .env and restart, or you’ll get a 421 Misdirected Request error. See Local & LAN deployment for details.

​
Agent skills

The MCP tools are deliberately generic — nothing in plamotrack knows what an “HG” grade is or how Bandai kit numbers work. You can teach your AI assistant those conventions by installing an agent skill: a small instruction file that loads alongside your MCP connection and keeps records consistent. The plamotrack-gunpla skill covers kit naming, grade buckets, series naming, Gundam Markers, and more. See the Agent Skills page for how to install one.