MCP Server

An AI assistant that answers a question about KNIME from memory often gets it wrong: it invents a node that does not exist, names the wrong extension, or describes a version that is two years old. The NodePit MCP server fixes that. It lets an assistant such as Claude, ChatGPT, or Cursor look things up on NodePit while it answers, so that it works with the actual nodes, extensions, and workflows instead of guessing.

MCP (the Model Context Protocol) is an open standard for connecting AI assistants to external tools and data. The NodePit server is free and needs no account or key. Its tools read public data and change nothing; the one exception, send_feedback, sends the NodePit team a note. Its address is:

https://nodepit.com/mcp

The MCP page on nodepit.com has the same setup steps and a list of its tools, which always matches the server.

Connect your assistant

Every assistant needs the same address. Where you enter it depends on the assistant. None of them asks you to sign in.

Claude

1. Go to Customize → Connectors and choose Add custom connector.
2. Paste https://nodepit.com/mcp and click Add.
3. In a chat, click + and switch NodePit on under Connectors.

Custom connectors work on every Claude plan; the free plan allows one. On Team and Enterprise plans, an owner adds the connector first under Organization settings, and then each member connects it for themselves.

Claude Code

Run this command in your terminal:

claude mcp add --transport http nodepit https://nodepit.com/mcp

The server is then available in the current project. Add --scope user to have it in every project.

ChatGPT

1. Open Plugins in ChatGPT on the web, click +, and choose Add custom MCP server.
2. Paste https://nodepit.com/mcp as the URL, set the authentication to No authentication, and confirm the warning.
3. In a new chat, type @ and pick the plugin.

Availability depends on your plan and your workspace settings, and OpenAI changes these menus now and then. OpenAI’s guide has the current steps.

Codex, Cursor, and VS Code

For Codex, run:

codex mcp add nodepit --url https://nodepit.com/mcp

For Cursor, add this to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for all projects:

{
  "mcpServers": {
    "nodepit": {
      "url": "https://nodepit.com/mcp"
    }
  }
}

For VS Code, add this to .vscode/mcp.json, or run MCP: Add Server from the command palette:

{
  "servers": {
    "nodepit": {
      "type": "http",
      "url": "https://nodepit.com/mcp"
    }
  }
}

Other assistants

Any assistant that speaks MCP over streamable HTTP works. If yours can configure itself, give it this prompt:

Add the remote MCP server "nodepit" to your configuration. It uses the streamable HTTP transport at https://nodepit.com/mcp and needs no authentication.

What you can ask

Describe what you need in your own words, as you would ask a colleague. The assistant decides by itself when to look something up. If it answers from memory, ask it to check NodePit. A few examples:

• “Which KNIME nodes can read Parquet files, and which extension do I have to install for them?”
• “Which node do people usually connect after a GroupBy node?”
• “Is the Row Filter node deprecated? If so, what replaces it?”
• “What does the QR Code Designer node do, and what do I have to install for it?”
• “Is there a public KNIME workflow I can start from for text mining?”
• “What is the latest KNIME release, and when did it come out?”

The guides show more, each with the request for the API and the question for an assistant.

What the server offers

The assistant picks from the tools below by itself. You never call them yourself, but their names show up in the assistant’s replies when it tells you what it did.

Tool	What the assistant uses it for
search	Finding nodes, categories, extensions, workflows, products, and users by name
get_node	A node in detail: description, ports, options, extension, deprecation, and the nodes most often used around it
list_nodes	A list of nodes, when no name is known
get_category	A category of the node repository, with the categories above it
list_categories	The categories of the node repository
get_extension	An extension and the update site to install it from
get_workflow	A public workflow or component, with the nodes it uses and a download link
list_workflows	Public workflows and components
get_product	A product by an extension vendor, with its latest version and its extensions
list_products	The products listed on NodePit
list_knime_versions	The released KNIME versions with their release dates, newest last
send_feedback	Reporting a tool that failed, wrong or missing data, or a missing capability to the NodePit team

Most of them accept a KNIME version, so you can ask for what is true for the version you actually use: “Does the Table Creator node in KNIME 5.3 have the same options?” Without a version, the assistant gets the latest one.

What the server sees

The server learns only what the assistant asks it: a search for “parquet”, or the ID of a node. It does not see your conversation, your files, or the other connectors you use, and there is no sign-in, so it has no account to link a request to.

The one tool that sends something is send_feedback. When a tool fails or something is missing, your assistant may report it to the NodePit team. NodePit then receives the text the assistant writes, its category, the tool it is about, the name of your client, and the address of the request. A person reads each report, and each client can send at most five per hour. Nothing the assistant does through the server changes any data on NodePit.

Limits

The server is free and shared, so each client can send 60 requests per minute. A question takes a handful of lookups, so you will rarely notice. If you do hit the limit, the assistant reports an error, and a minute later it works again. The MCP page always shows the current limit.

Troubleshooting

The assistant cannot connect. Check that the address is exactly https://nodepit.com/mcp, with no trailing characters, and that the authentication is set to none where the assistant asks for it, as with No authentication in ChatGPT. The server does not use any, and an assistant that waits for a sign-in will not connect.

The assistant does not look anything up. Some assistants only use a connector if it is switched on for the chat (in Claude, under Connectors next to the message box) or if you mention it. Ask it to “check NodePit” and it will.

The answer says “Not found”. The ID the assistant used does not exist, often because it guessed one, or the item is not part of the KNIME version it asked for. Ask it to search for the item by name first.

The answer looks out of date. Ask which KNIME version the answer is for. NodePit answers for the latest version unless a version is given.

Using the REST API instead

The MCP server and the REST API read the same data. Use the server when an assistant should do the looking, and the API when you write the code yourself. The server trims its answers to what fits into a conversation: a node comes with its ten most common predecessors and successors instead of all of them, and the KNIME versions come once per release instead of once per platform. The API has the complete data, paging, and the node images.

NodePit Runner has an MCP server of its own, for managing your projects, schedules, and runs. It is a separate server, and you can connect both to the same assistant. See AI Assistant (MCP Server) for Runner’s.