Docs MCP Server — Search These Docs from Your AI Tools

This is an MCP (Model Context Protocol) server that lets AI tools find and read the PlantPulse documentation directly. Add one URL to any MCP-capable tool — Claude Code, Claude Desktop, Cursor, VS Code — and during a conversation it can search the docs, read whole pages, and answer with links to its sources.

Anyone can use it. No sign-up and no API key are required. It is read-only: it cannot change the documentation or reach any other system.

Connection Details

Item	Value
URL	`https://docs.plantpulse.io/mcp`
Transport	Streamable HTTP (stateless — no sessions)
Authentication	None
Languages	`en` (default) · `ko` · `ja` · `zh` · `de`
Coverage	Everything currently published on this site — Platform · Edge · AI · Studio · Resources

When the documentation is updated and the site is redeployed, the MCP server's content changes with it. There is no separate index to wait for.

Connecting

Claude Code

claude mcp add --transport http plantpulse-docs https://docs.plantpulse.io/mcp

After adding it, claude mcp list should show plantpulse-docs as ✓ Connected. To share it with everyone working on a project, add --scope project (it is written to .mcp.json).

Claude Desktop

Under Settings → Connectors → Add custom connector, enter the URL https://docs.plantpulse.io/mcp. Leave the authentication fields empty.

To configure it in the settings file (claude_desktop_config.json) instead, use mcp-remote, which bridges to remote servers (requires Node.js).

{
  "mcpServers": {
    "plantpulse-docs": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://docs.plantpulse.io/mcp"]
    }
  }
}

Cursor

Add it to ~/.cursor/mcp.json (all projects) or to the project's .cursor/mcp.json.

{
  "mcpServers": {
    "plantpulse-docs": {
      "url": "https://docs.plantpulse.io/mcp"
    }
  }
}

VS Code (GitHub Copilot agent mode)

Add it to .vscode/mcp.json in the workspace.

{
  "servers": {
    "plantpulse-docs": {
      "type": "http",
      "url": "https://docs.plantpulse.io/mcp"
    }
  }
}

ChatGPT

In ChatGPT, connect it as a custom connector (custom MCP server).

Open the screen for adding a custom MCP server. Currently this is Plugins (chatgpt.com/plugins) + → Add custom MCP server; depending on the ChatGPT version it may instead be Settings → Connectors (Apps & Connectors) → Advanced settings, where you turn on Developer mode and then choose Create.
Enter a name (for example PlantPulse Docs) and a description, and enter https://docs.plantpulse.io/mcp as the URL.
For authentication, choose None (No authentication).
Read the risk notice, accept it and create the connector. You are done when the tool list reported by the server shows search · fetch.

Whether this menu is available depends on your ChatGPT plan and workspace settings (whether an admin allows custom connectors), and the menu names change between ChatGPT versions. ChatGPT connectors and deep research look for tools named exactly search · fetch, so this server provides those two tools separately (see Tools below).

Other MCP Clients

Most clients that support Streamable HTTP accept the following shape.

{
  "mcpServers": {
    "plantpulse-docs": {
      "type": "http",
      "url": "https://docs.plantpulse.io/mcp"
    }
  }
}

Tools

All five are read-only.

Tool	Arguments	Returns
`search_docs`	`query` (required, up to 300 characters) · `product` · `locale` · `limit` (1–20, default 8)	The best-matching sections — title · section heading · product · URL · excerpt
`get_page`	`path` (required) · `locale`	One page's title · URL · section list · full body (up to 60,000 characters)
`list_pages`	`product` · `locale` · `cursor` · `limit` (1–200, default 100)	A list of pages — path · title · product. Fetch the next batch with `cursor`
`search`	`query` (required) · `locale` · `product` · `limit` (1–20, default 8)	For ChatGPT. `{"results":[{id, title, url, text}]}` — one result per page
`fetch`	`id` (required) — an id returned by `search`, or a docs URL	For ChatGPT. `{id, title, text, url, metadata}` — the full body of one page

product is one of platform · edge · ai · studio · resources. Several products have pages with the same title, such as “Changing passwords and API keys”, so set it when the question is about a specific product.
Without locale, searches run in English (en). For questions in Korean, ko works best.
The path of get_page accepts a site path such as /platform/frontline/user-guide, a path with a language prefix such as /ko/..., or a full https://docs.plantpulse.io/... URL. For unknown paths it suggests similar pages.
search · fetch return the same index in the names and shape ChatGPT expects. Without locale, search picks the language from the characters of the query (Hangul → ko, kana → ja, other CJK → zh, German → de, otherwise en). An id is a page path with its language prefix, such as /ko/edge/user-guide/plc-profiling-testkit.
Search uses the same index and method as this site's docs chat (keyword BM25, including handling of Korean particles).

The server also tells the AI tool to “cite the returned URLs when answering”. When an answer carries links, you can open those pages to check the original text.

Usage Limits

Because the server is public, a few limits keep any one user from monopolizing it.

Item	Limit	When exceeded
Request rate	About 30 per minute per IP (bursts of up to 20 allowed)	`429` — retry after a minute
Request size	64 KB	`413`
Query length	300 characters	Tool error
Search results	20 per call, excerpts of about 700 characters	Truncated
Page body	60,000 characters	Truncated, with the source URL appended

An AI tool typically uses three or four requests to connect (initialization · tool list) and a few more searches and page reads per question. Normal use stays well within the limits.

What is logged

The server keeps only the same web server access log as the docs site (client IP · time · URL). Search queries and tool arguments are not stored. Do not put confidential company text into queries — this is a public server.

Trying It Yourself

You can see it working with curl, without any client. The Accept header must list both formats.

curl -s https://docs.plantpulse.io/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_docs","arguments":{"query":"PLC profiling","locale":"en","limit":3}}}'

Server status (version · page counts) is at https://docs.plantpulse.io/mcp/healthz.

Troubleshooting

Symptom	Cause and fix
Connection fails · `406`	The client does not support Streamable HTTP, or the `Accept` header is missing. Clients that only support the older SSE transport can connect through `mcp-remote`, as with Claude Desktop.
`405`	The request used `GET`. This server is stateless and accepts only `POST`.
`429`	Too many requests. Retry after about a minute.
`503`	The site was just redeployed and the server is loading its index. Retry in a few seconds.
Results only in English	No `locale` was given. Ask the AI tool to “search the Korean docs” and it will call with `ko`.

Connection Details​

Connecting​

Claude Code​

Claude Desktop​

Cursor​

VS Code (GitHub Copilot agent mode)​

ChatGPT​

Other MCP Clients​

Tools​

Usage Limits​

Trying It Yourself​

Troubleshooting​