MCP server

MCP (Model Context Protocol) is a standard interface that AI clients, such as Claude or ChatGPT, use to connect to an application's data and actions as a set of tools. TAS exposes its own MCP server at the address of your installation. Through it, an AI client can create cases, look them up, edit task variables or read company documentation – always within the permissions of the user who connected to the server. This chapter is intended for administrators who allow AI clients to access TAS and for integrators who connect the client.

Address and scenarios

The MCP server listens on https://vase-tas.cz/mcp, optionally with a scenario name in the path:

https://vase-tas.cz/mcp/case-creation

A scenario narrows the set of tools offered to what the given AI client is meant to do. Without a scenario (/mcp), the AI client sees all tools the user has permission to use.

ScenarioPurpose
case-creationCreating a new case: find templates/headers that can be started, determine the required variables, validate the input and create the case. Also includes tools for searching the company AI documents.
case-editingEditing a case in progress: looking up the case, triggering manual case events, editing variables of a running task.
case-lookupOnly looking up and listing the user's cases through their views (Custom Views).
knowledgeSearching the company AI documents and the TAS documentation (HelpDocs).
accountRoles of the signed-in user and their previous conversations with the AI.
administrationOverview of installed plugins. Administrators only.
(no scenario)All of the tools above plus tools for working with the currently open case (overview, documents, variables, tasks, history, events).

TAS communicates using the MCP protocol over HTTP (JSON-RPC in the body of a POST request), without keeping state on the server side and without streaming over SSE – the GET and DELETE methods on the same address return "Method not allowed".

Authentication and connecting an AI client

The MCP server accepts only an API token or an OAuth token, not a regular browser sign-in session. For an AI client used by a specific user (the typical case), use OAuth with an authorization code – the procedure is the same as for any public OAuth application:

  1. An administrator with the $ApiTokenAdmin role creates a public OAuth client in API Access and enters the Redirect URI required by the AI client (see the article API tokens and OAuth clients in practice).
  2. In the AI client, enter the MCP server address (https://vase-tas.cz/mcp, optionally with a scenario) and the Client ID.
  3. The client redirects the user to /oauth/authorize; TAS displays the Application Authorization screen with the application name and redirect address, and the user approves or denies access.
  4. The client exchanges the code for an access token at /oauth/token and from then on calls MCP with the Authorization: Bearer <token> header.

For an integration without user involvement (e.g. a company automation running as a single system account), you can use an API token or a confidential OAuth client instead – the MCP server accepts both types of tokens in the same way.

Permissions follow the connected user
The AI client can see and do only what the user (or system user) represented by the token is allowed to. Tools marked as write operations (creating a case, editing variables, triggering an event) may additionally require the AI client's own confirmation before they are executed.

Available tools

ToolsWhat they do
Case creation (case-creation)Find templates and headers the user is allowed to start, return the required variables as a schema, validate the filled-in data and create the case.
Documents and knowledge (AiDocumentsTools, HelpDocsTools)Search the organization's company AI documents (first by keywords, then the whole document) and the TAS documentation.
Case lookup (case-lookup, case-editing)Go through the views (Custom Views) available to the user and find a specific case in them.
Case editing (case-editing)Determine the available manual case events and trigger the selected one; determine the variables of a task in progress as a schema, validate and save their new value.
Roles and conversations (account)Return the roles of the signed-in user and their previous conversations with the AI assistant.
Plugins (administration)Administrators only: overview of installed plugins and their properties.
Open case (no scenario)Overview, documents, variables, active tasks, history and manual events of the currently open case.

Each group of tools starts with a tool that explains to the AI client how to search or filter further (e.g. which columns a view has), followed by the tool that actually returns or writes the data – this way the AI client verifies the context first before performing an action.

Where to find out what the AI is allowed to do
The specific list of roles and permissions the AI client inherits is set on the user or system user in the same way as for a regular sign-in – in Users and Roles. The MCP server neither adds nor removes any permissions of its own.

Related: API Access, API tokens and OAuth clients in practice, Authentication.

MCP server
Syca (opens in a new tab)

Powered by HelpDocs (opens in a new tab)