Vertec MCP Server

As of version 6.8.0.21, Vertec offers the Vertec MCP Server.

MCP (Model Context Protocol) is an open standard for connecting AI applications to external systems. With the Vertec MCP Server, AI clients such as Claude or ChatGPT can connect to Vertec and both query structured data and perform tasks in Vertec such as creating new objects.

The Vertec MCP Server runs together with every Vertec instance in the Vertec Cloud Server. The MCP itself is an intermediate layer that publishes tools (and the texts) against external tools, receives requests, and forwards them internally to the Vertec web services. It does not do anything that could not be done via the REST API, nor is it faster in processing data. It is not suitable for processing large lists of objects, and if you want to analyze measures with it, licensing the Business Intelligence module is strongly recommended. The difference is that it allows you to connect structured Vertec data (including customizing classes, custom field items and link types) to an AI system and analyze it using AI.

The Vertec MCP Server has knowledge of the Vertec Model, which makes it possible to query structured data in Vertec. This means that queries are based on the model, and not based on text content as in an LLM / RAG system. The graph below shows the interaction of the different components:

Description of the graphic in steps

1. The user asks a question or a request in the chat. The AI client accepts the request and coordinates the process.
2. The request goes on to the LLM and that realizes that up-to-current information is needed that is not available in his “training knowledge”.
3. The LLM specifies on the AI client which tool should be used for this purpose.
4. The AI client receives the prompt and searches for the tool in the Vertec MCP server.
5. The Vertec MCP Server provides the tools and serves as an intermediate layer to the Vertec web services.
6. The actual search of the data takes place via the two web services REST and ModelMetdata.
7. The results found from the Vertec object system are returned to the AI client and forwarded to the LLM.
8. The LLM creates the response from the processed information for the user in the chat.

The tools available in the Vertec MCP Server are connected via so-called MCP connectors. See the section Connect to vertec.

Commissioning

The Vertec MCP Server is part of the Vertec Cloud Servers and basically runs when the Cloud Server is running, but it is not really started until the first access is made to save resources when not in use. When running Vertec On-Premises, the MCP Server can also be turned off completely in the Vertec.ini – file.

The Vertec MCP Server’s knowledge of the Vertec Model is controlled by the Vertec web service called ModelMetadata. This must be aktiv in order to work with the Vertec MCP Server. The requests to the data are made internally by the Vertec MCP Server via the Vertec rest api. Therefore the REST web service must be aktiv.

Authentication of the Vertec MCP Server to Vertec is done via OAuth2, the Vertec Cloud Server acts as an OAuth server. For this purpose, the following parameters are available in the system settings Authentication:

• The Oauth2 client id is created automatically by Vertec and is already registered.
  • The OAuth2 flow in Vertec supports PKCE if access from the external system is done this way (if the client sends a code_challenge parameter in the authorize request).
• Under Oauth2 client redirect url, you must specify the OAuth callback URLs of the external systems to which you want to allow a request (comma-separated). These must be HTTPS addresses, otherwise a warning will be issued when connecting.
  • Claude’s redirect URL is: https://claude.ai/api/mcp/auth_callback.
  • In Chatgpt, your own server is part of the redirect URL, so the redirect URL can only be read after registering the Vertec MCP server in Chatgpt. It can be found in App registration (plugin) > Advanced OAuth settings > Callback URL. This is only shown if the registration method is Custom OAuth client.

    

• You can also activate Dynamic client registration (dcr)

Standard values

The Vertec MCP Server comes with the following default values, so users can get started straight away:

• The Standard Users group has granted accepted MCP user rights by default. assign the MCP permission to the Standard Users group.
• The ModelMetadata and REST web services are active by default.
• The OAuth2 Dynamic Client Registration (DCR) system setting is activated.
• The Claude Callback URL is pre-entered as the Oauth2 Client Redirect URL.

Connect to Vertec

The external connection is via the URL where your Vertec can be reached from the outside, followed by /mcp, e.g.:

https://meinefirma.vertec-cloud.com/mcp

This URL is stored in the external system as a data source, together with the OAuth2 Client ID or generated via Dynamic client registration (dcr) via the endpoint /register.

• For Chatgpt, the registration URL must first be specified in the app registration under Advanced OAuth settings, e.g. https://meinefirma.vertec-cloud.com/register. Only then can DCR be selected as the registration method above.

When using the AI client, the above information is entered as connectors. For this purpose, refer to the following article on custom connector.

Then the connection to Vertec is established.

If the user is not yet logged in to Vertec, the user is now logged in as usual. The OAuth flow always takes you to a login screen, even if the user is already logged in to Vertec in the same browser. The current session is not used.

In order to be able to make requests to the Vertec MCP server, it must have the User Right user rights in Vertec. This does not grant any additional rights, the same rights apply as when accessed via a UI.

a message is shown informing the user about the user rights to be granted to the external system:

This can be confirmed or rejected:

• Upon confirmation, the OAuth flow continues normally, i.e. a token is issued and the user can begin the requests.
• If rejected, no token is issued and the client receives an error.
• If the OAuth2 Client Redirect URL is not HTTPS and does not match the system setting, the following warning is displayed:

The Vertec model

The Vertec MCP Server has a reduced Vertec model available, so it does not export the entire model as shown in the Vertec Model Browser, but only the most important classes for a data query.

In addition, all classes for which there is an entry in the Class Settings are also taken into account, since we assume that they are relevant to the customers. It is important to include entering descriptions in the comment field in the class settings, custom field items and custom link types that you created yourself so that the AI knows what the request is about.

This is controlled by the Vertec Web Service called ModelMetadata, where the corresponding code can also be viewed. Important classes that are on the whitelist are exported, as well as all classes for which class settings exist.

Requests to the Vertec MCP Server

As soon as an external system such as Claude, OpenAI or ChatGPT is connected to the Vertec MCP server, queries can be made to Vertec in the usual text form. Datum can be queried as well as new objects can be created in Vertec.

Example of a request to Vertec (with licensed Business Intelligence module) with Anthropic Claude:

Further insight into the possibilities of communication with the Vertec MCP Server can be found in the following video:

Using vertec erp & crm with an mcp server