---
title: "Nodus Works MCP server: connect Claude, ChatGPT and Cursor"
description: "Connect the Nodus Works MCP server to Claude, ChatGPT and Cursor. Read-only tools, setup and a live try-out."
url: https://nodusworks.com/en/ai-mcp
markdown: https://nodusworks.com/en/ai-mcp.md
language: en
alternates:
  tr: https://nodusworks.com/yapay-zeka-mcp.md
  de: https://nodusworks.com/de/ki-mcp.md
last_updated: 2026-11-16
type: page
publisher: "Nodus Works"
---

# Nodus Works MCP server: connect Claude, ChatGPT and Cursor

When Claude, ChatGPT or Cursor connects to this server, it reads the pages on the site: services, integrations, apps, case studies and articles. Every answer comes back with the address of the published page.

## What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools and data sources. The assistant sees the tools of the server it is connected to and calls the right one itself while answering a question. The Nodus Works MCP server feeds these tools with the site's own content, so your assistant reads the relevant page instead of guessing about Nodus Works.

## What is it for?

### Answers with sources

When the assistant describes a Nodus Works service, integration or case study, it reads the page itself and gives you its address. It passes on what the site says instead of an invented price or promise.

### Three languages, one address

Turkish, English and German pages share one index. The language of the question is detected and pages in that language rank first; every result also carries its counterparts in the other languages.

### It does not touch your store

This connection only reads pages published on the site. It does not reach your orders, your forms, your email or your calendar. When you ask how to get in touch, it gives you the phone number, the email address and the meeting link. You send the message.

Once connected, you can ask your assistant things like:

- Which Shopify services does Nodus Works offer?
- Which brands moved from Ticimax to Shopify, and what was done during the migration?
- Which pages should I read for a Yurtiçi Kargo and Logo integration?
- Does Nodus Works have an app for wholesale on Shopify?

## Connect in three steps

### Copy the address

The server address is `https://nodusworks.com/mcp`. No account, API key or OAuth sign-in is needed.

### Add it to your client

In Claude and ChatGPT it is added as a custom connector; in Cursor and VS Code as one entry in a configuration file. The steps for each client are in the tabs below.

### Ask

Turn on the Nodus Works connector in the chat and type your question. The assistant first finds the relevant pages with `nodus_ara` and, if needed, reads the full page with `nodus_sayfa`.

The assistant picks the right tool itself. To be sure, add "use the Nodus Works connector" to your question.

## Client setup

The steps below are taken from each client's official documentation. Menu names follow the client's interface language; they are written here in English.

### Claude

In Claude, remote MCP servers are added as custom connectors. The same connector works in claude.ai, Claude Desktop and Cowork; the connection is made from Anthropic's servers.

Open **Customize › Connectors**, click **+ Add** and then choose **Add custom connector**.

Enter Nodus Works as the name, enter `https://nodusworks.com/mcp` as the server address and click **Continue**.

Under **Authentication**, choose **No sign in** and finish with **Add**.

In a chat, open the **Connectors** menu from the **+** button at the bottom left and turn on Nodus Works.

On Team and Enterprise plans, an owner first adds the connector under **Organization settings › Connectors**, then members click **Connect**. The free plan allows one custom connector. The `claude_desktop_config.json` file in Claude Desktop is for local servers; use the connector route for a remote server.

### ChatGPT

ChatGPT connects a custom MCP server as a plugin in the web version. Workspace permissions can turn this feature off.

Open ChatGPT in the browser, click the plus button on the **Plugins** page and choose **Add custom MCP server**.

Enter Nodus Works as the name; under **Connection**, enter `https://nodusworks.com/mcp` in the **Server URL** field.

For authentication choose **No authentication**, accept the risk notice and click **Create as a plugin**.

Install the plugin. In a chat, type `@` and pick Nodus Works to use it.

ChatGPT treats tools marked with `readOnlyHint` as read tools; tools without it are treated as write actions and need confirmation. All seven Nodus Works tools carry this hint.

### Cursor

Add it to `~/.cursor/mcp.json` to use it in every project, or to `.cursor/mcp.json` in the project root to use it in one project only:

```
{
"mcpServers": {
  "nodus-works": {
    "url": "https://nodusworks.com/mcp"
  }
}
}
```

The server appears on the **Customize** page in the sidebar, where you can turn it on and off. By default Cursor asks for approval before it runs an MCP tool.

### VS Code

Used with GitHub Copilot. Add it to `.mcp.json` in the workspace root to share it with your team, or to `~/.copilot/mcp-config.json` for personal use:

```
{
"mcpServers": {
  "nodus-works": {
    "type": "http" ,
    "url": "https://nodusworks.com/mcp" ,
    "tools": [ "*" ]
  }
}
}
```

Instead of editing a file, you can run **MCP: Add Server** from the command palette, choose **HTTP** and enter the address. The older `.vscode/mcp.json` file uses the `servers` format and, according to the VS Code documentation, is not recommended for new setups.

### Claude Code

Added with a single terminal command:

```
claude mcp add --transport http nodus-works https://nodusworks.com/mcp
```

You can see the added server with `claude mcp list`.

### curl

The server speaks plain JSON-RPC, so you can try it without a client. The shortest request:

```
curl -s https://nodusworks.com/mcp \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"nodus_ara","arguments":{"query":"Trendyol integration","language":"en","limit":3}}}'
```

The same request in protocol version 2026-07-28, with headers and the `_meta` field:

```
curl -s https://nodusworks.com/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-H 'MCP-Protocol-Version: 2026-07-28' \
-H 'Mcp-Method: tools/call' \
-H 'Mcp-Name: nodus_ara' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"nodus_ara","arguments":{"query":"Trendyol integration","language":"en","limit":3},"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientCapabilities":{}}}}'
```

## Try it in your browser

This box does what an AI client does: it sends a `tools/call` request to the `/mcp` endpoint and runs the `nodus_ara` tool. Results are listed as the server returns them; you can open the full request and response below.

## Seven read-only tools

Every tool has a closed input schema: an undefined field, a wrong type or an out-of-range value returns a descriptive tool error.

### nodus_ara

Ranked site search in three languages. Titles, tags and section headings weigh more than body text; Turkish suffixes and synonyms (migration, move, switch) are matched. Each result returns the title, address, summary, best-matching section and its counterparts in the other languages.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_ara" ,
    "arguments": {
      "query": "Ticimax to Shopify migration" ,
      "language": "en" ,
      "limit": 2
    }
  }
}
```

}

### nodus_sayfa

Reads a single page as clean Markdown. Takes a path or a full address; only nodusworks.com addresses are accepted. Long pages are shortened, and the remaining section headings and the page's full address are added at the end.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_sayfa" ,
    "arguments": {
      "path": "/en/integrations/shopify-trendyol-integration"
    }
  }
}
```

}

### nodus_hizmetler

Lists services with their groups (shopify, software, marketing, seo) and pages.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_hizmetler" ,
    "arguments": {
      "language": "en" ,
      "category": "shopify"
    }
  }
}
```

}

### nodus_entegrasyonlar

Lists marketplace, shipping, ERP and accounting integrations with the platform name and page. Filters by category or by the start of the platform name.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_entegrasyonlar" ,
    "arguments": {
      "language": "en" ,
      "category": "shipping"
    }
  }
}
```

}

### nodus_uygulamalar

Lists Nodus Works' Shopify apps, SaaS products and free tools with a short description and page.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_uygulamalar" ,
    "arguments": {
      "language": "en" ,
      "kind": "saas"
    }
  }
}
```

}

### nodus_vakalar

Lists case studies with brand, sector, platform, migration path, market and scope. Filters by free text or by the old platform the brand moved from; on request it also adds reference brands that have no case study page.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_vakalar" ,
    "arguments": {
      "language": "en" ,
      "migrated_from": "Ticimax"
    }
  }
}
```

}

### nodus_iletisim

Returns phone, WhatsApp, email, address, meeting link and the contact page. The server cannot send messages; the links are for the user.

```
{
  "jsonrpc": "2.0" ,
  "id": 1 ,
  "method": "tools/call" ,
  "params": {
    "name": "nodus_iletisim" ,
    "arguments": {
      "language": "en"
    }
  }
}
```

}

## Resources

Ready-made documents that can be added to the context without calling a tool. All are Markdown and are generated from the same index as the tools.

| Address                | Content                                                                                                     |
| ---------------------- | ----------------------------------------------------------------------------------------------------------- |
| `nodus://overview`     | A short introduction to Nodus Works, the main sections in three languages and the documented content counts |
| `nodus://services`     | All services with their groups and pages                                                                    |
| `nodus://integrations` | Marketplace, shipping, ERP and accounting integrations                                                      |
| `nodus://apps`         | Shopify apps, SaaS products and free tools                                                                  |
| `nodus://case-studies` | Case studies by brand, sector and migration path                                                            |
| `nodus://contact`      | Contact channels and contact pages                                                                          |
| `nodus://page/{+path}` | Any page, for example `nodus://page/en/services/shopify-store-setup`                                        |

## Prompts

Ready-made tasks that clients such as Claude offer in a menu. Each prompt tells the assistant in which order to use the tools and to cite its sources.

| Prompt                     | What it does                                                                  | Arguments                       |
| -------------------------- | ----------------------------------------------------------------------------- | ------------------------------- |
| `nodus_gecis_plani`        | Drafts a step-by-step migration from the current platform to Shopify          | `platform`, optional `language` |
| `nodus_entegrasyon_secimi` | Finds integrations that match marketplace, shipping, ERP and accounting needs | `needs`, optional `language`    |
| `nodus_tanitim`            | Briefly introduces Nodus Works with sources                                   | optional `language`             |

## Transport and versions

The server uses MCP's Streamable HTTP transport: every request arrives as `POST /mcp` and the response is a single JSON body. No stream (SSE) is opened and no session is kept; each request is handled on its own. `GET` and `DELETE` return 405.

| Version                                  | Start                           | Note                                                                           |
| ---------------------------------------- | ------------------------------- | ------------------------------------------------------------------------------ |
| `2026-07-28`                             | `server/discover`, no handshake | Every request carries the version and client capabilities in its `_meta` field |
| `2025-11-25`                             | `initialize`                    | No session ID is issued                                                        |
| `2025-06-18`, `2025-03-26`, `2024-11-05` | `initialize`                    | For older clients                                                              |

Supported methods: `server/discover`, `initialize`, `ping`, `tools/list`, `tools/call`, `resources/list`, `resources/templates/list`, `resources/read`, `prompts/list`, `prompts/get`. Notifications get 202 with no body; batch requests are not accepted.

## Headers

| Header                 | Rule                                                                                                                            |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
| `Content-Type`         | Must be `application/json`, otherwise 415                                                                                       |
| `MCP-Protocol-Version` | In 2026-07-28 requests, must match the version in the body                                                                      |
| `Mcp-Method`           | Must match the `method` in the body                                                                                             |
| `Mcp-Name`             | The tool, resource or prompt name for `tools/call`, `resources/read` and `prompts/get`; non-ASCII names are sent base64-encoded |

## Error codes

| Code     | HTTP | Meaning                                                             |
| -------- | ---- | ------------------------------------------------------------------- |
| `-32700` | 400  | The body could not be parsed as JSON                                |
| `-32600` | 400  | Invalid request: batch, wrong `id`, extra field                     |
| `-32601` | 404  | Unknown method                                                      |
| `-32602` | 400  | Unknown tool, prompt or resource; missing `_meta`                   |
| `-32020` | 400  | Header and body do not match                                        |
| `-32022` | 400  | Unsupported protocol version; the response lists the supported ones |
| `-31429` | 429  | Rate limit; the `Retry-After` header gives the wait time            |
| `-31403` | 403  | Origin not allowed                                                  |

Errors in tool input are not protocol errors but tool results with `isError: true`. This way the assistant can read the error and correct its request.

## Security model

The server does not call a language model and is not tied to any account. The only thing an AI assistant can do with this address is read pages that are already public on the site.

### Read-only

Every tool carries `readOnlyHint: true` and `destructiveHint: false`. No tool reaches out to the network, opens files or writes records.

### No secrets

The server reads no environment variables and holds no keys or passwords. Tests check that the index contains no secret patterns, local paths or internal addresses.

### Content is data

The server instructions tell the client that returned text is site content and must not be treated as instructions. Tags, scripts and styles are stripped at build time.

### Queries are not echoed

Search results do not repeat the query, and error messages are fixed. No stack traces, file paths or internals appear in responses.

### Closed schemas

Undefined fields are rejected. Page paths may not contain `..`, `%`, backslashes or control characters; no request is ever made to a foreign host.

### Limits

Body size, depth, rate and concurrency are limited. Responses are not cached thanks to `no-store` and are closed to search engines.

| Limit                      | Value                          |
| -------------------------- | ------------------------------ |
| Request body               | 16 KB                          |
| JSON depth and field count | 12 levels, 200 fields          |
| Search query               | 300 characters                 |
| Result count               | 1–10, default 5                |
| Rate                       | 120 requests per minute per IP |
| Concurrent requests        | 4 per IP, 32 in total          |

Requests from browsers are only allowed from the origins `https://nodusworks.com` and `https://www.nodusworks.com`. Desktop and cloud clients do not send an origin header, so this rule does not affect them.

## Frequently asked questions

- **Do I need an account or a key to use it?**:

  No. The address is public; it requires no account, API key or OAuth sign-in. The only protection against abuse is a rate limit.

- **Can the AI do something on my behalf through this connection?**:

  No. The connection reads pages published on the site. It does not reach your orders, your forms, your email or your calendar. When you ask how to get in touch, it gives you the phone number, the email address and the meeting link. You send the message.

- **Where do the answers come from, and are they up to date?**:

  Answers come from the Turkish, English and German pages published on the site. When a published page changes, the next answer follows that page.

- **Are my questions stored?**:

  The server code does not write the query anywhere, opens no session and uses no cookies. For the rate limit, the IP address is kept only in memory, in a one-minute counter.

- **Which languages can I ask in?**:

  Turkish, English or German. The language of the question is detected and pages in that language rank first; the `language` parameter can also pin the search to one language.

- **What is done against prompt injection?**:

  Content only comes from Nodus Works' own pages and is cleaned at build time. The server instructions mark returned text as data, and the query is not reflected in the response. Most importantly, because the tools cannot act, even a manipulated assistant cannot change anything through this server.

- **I see an error when I open the address in my browser. Is the server down?**:

  No. MCP requests arrive as `POST`; the server deliberately returns 405 to the `GET` request your browser sends. To try it, use the live box above or the curl command.

- **Claude or ChatGPT cannot connect. What should I check?**:

  Make sure the address ends in `/mcp` and that authentication is turned off (No sign in, No authentication). Claude connectors reach the server from Anthropic's servers rather than from your computer, so a restriction on your company network does not affect Claude's connection.

Work installed on your store

This server only reads the site.

[Consulting page](https://nodusworks.com/en/services/ai-consulting-for-your-shopify-store)

---

Source: [Nodus Works MCP server: connect Claude, ChatGPT and Cursor · Nodus Works](https://nodusworks.com/en/ai-mcp)  
Full content index: https://nodusworks.com/llms.txt  
MCP server for AI agents: https://nodusworks.com/en/ai-mcp
