Native MCP Server for Label Printing Agents

Connect over Streamable HTTP

Use an MCP client that supports Streamable HTTP and a configured authorization header. Connect to the Center, using your installation's HTTPS address.

Endpoint: https://center.example/mcp
Transport: Streamable HTTP; stateless, with JSON responses.
Authorization: Authorization: Bearer <API-TOKEN> on requests.

The endpoint is exactly /mcp. Browser session cookies are not accepted there. Let your MCP client perform initialization, then discover the available tools and resources. The Runner has no MCP endpoint.

Create an agent identity

An administrator can issue a token in the Center's API-token settings or with POST /api/v1/auth/tokens. This example creates a document-authoring identity:

POST /api/v1/auth/tokens
Authorization: Bearer <ADMIN-TOKEN>
Content-Type: application/json

{"subject":"label-agent","role":"designer"}

The response returns the token once, in token. Store it in your client's secret configuration. Optional expires_at uses RFC 3339; omitting it creates a token without an expiry. Revoke a token with DELETE /api/v1/auth/tokens/{id}.

designer permits document authoring.
operator permits operational work, including access to Runner status.
approver supplies review permissions, subject to the human-approval policy.
admin permits administration, including credential management.

Roles are broad permission groups. For an observation-only agent, also restrict its available tools in the client; an operator token is not a dedicated read-only token. The server rechecks current roles for requests.

Discover tools from the running Center

tools/list returns the tools permitted by the identity's roles. The server generates tool names and input schemas from its API contract: a tool's name is the operation's operationId. Authorization is enforced again when a tool is called.

Path, query and header parameters are top-level tool arguments.
The JSON request body goes under body.
For file uploads, a file field contains {"name":"example.nlbl","contentBase64":"…"}.

Useful implemented tools include:

listLabelVorlagen, createLabelFromVorlage, patchLabelDocument and postLabelPreview for labels.
createFlow, patchFlowDocument, postFlowLint and postFlowDryRun for automations.
postLabelCopilot, postFlowCopilot and postFormCopilot for Center-hosted proposals.
listFlowRuns, listRunners, getRunner and listPrintArchive for operation.
aiLintErklaerung and aiLogQA for diagnostic explanations. Log Q&A accepts structured audit-history filters.

Read each returned input schema before calling it. Availability in the list reflects roles; approval policy and document validation can still reject an attempted action.

Read the schemas before constructing documents

Use resources/list and resources/read to obtain the contract from the connected Center:

ole://registry/labels — supported label object types.
ole://registry/flows — automation node definitions.
ole://schema/label, ole://schema/flow and ole://schema/form — document schemas.
ole://openapi — the API specification.

Preserve existing object and node IDs when changing a document. Use supported nodes, fields and CEL expressions to define behavior. MCP integrates Center capabilities into an agent workflow; custom integration code runs in your own environment.

A first label workflow

Call listLabelVorlagen and choose an actual template ID from the response.
Call createLabelFromVorlage with that ID and a name. Keep the returned label ID.
Read the document and schema. Propose changes, check them and use patchLabelDocument to save an intended revision.
Call postLabelPreview with sample values for the label's declared variables.
Have a person review and approve the agent-authored version under the default policy.

For example, these are the parameters an MCP client passes to tools/call to create the template copy:

{
  "name": "createLabelFromVorlage",
  "arguments": {
    "body": {
      "vorlageId": "<template-id-from-list>",
      "name": "Packing label"
    }
  }
}

A preview request for a label with declared partNumber and lot inputs could use:

{
  "name": "postLabelPreview",
  "arguments": {
    "labelId": "<label-id>",
    "stufe": "etikett",
    "body": {
      "variables": {
        "partNumber": "BRG-6204",
        "lot": "L-204"
      }
    }
  }
}

Use the variables your chosen label actually declares. A template may need other inputs to render. Preview returns image content; JSON responses return text. A second text block supplies HTTP status, content type and response headers such as X-Label-Dots. Failed operations return isError: true with the API error code and parameters.

For automations, use postFlowDryRun to inspect a draft with pinned sample inputs. Its result distinguishes executed logic, supplied source data and side effects that were not performed; it does not place physical print jobs.

Keep the review controls in the workflow

Every MCP request uses a bearer token and is treated as a service actor. With aiFreigabePflicht enabled, the default, agent-authored label and form versions remain drafts. New automation deployment and deployment approval require a human.

An administrator using a human session can change this policy; the change is audited. A token cannot disable the requirement itself. A rollback to an already deployed version has its own permitted operational path, so the policy should not be read as a ban on every operational action.

Copilot tools return proposals. They do not save the document. Apply accepted patches through the ordinary document tools and retain the version and author information for review.

The MCP catalog excludes browser login/setup, the credential-store master-key export and Runner bootstrap operations. Admin-only credential creation and deletion remain available through the usual role checks.

Configure observation and extend your agent's workflow

Schedule your external agent to inspect permitted run history, Runner status and print outcomes. Alternatively, configure an external receiver for supported outgoing webhook events. The event catalog includes print.failed, print.held, print.sent, print.confirmed and label.approved.

Use the event or poll result to collect relevant context, investigate a pattern and propose a document change. Keep sent and confirmed distinct in the agent's report. Combine these Center tools with tools for your own ERP, knowledge base or support workflow.

The agent runtime, schedule, receiver and notification destination are yours to configure. Runner online/offline events are not in the outgoing webhook catalog; use permitted status reads for that information. Built-in proactive monitoring and policy-driven autonomous fixes are planned.

Explore webhook integration →

Separate your agent's model from Center Copilots

An external agent can use document and operational MCP tools with its own model setup. Connecting MCP does not require a Center LLM credential.

The Center's Copilot and model explanation tools use a saved LLM credential. Choose anthropic for the Messages format or openai for OpenAI-compatible Chat Completions, specify model, and configure endpoint and apiKey as needed. Compatible local models can use the OpenAI format; remote endpoints require HTTPS, while local loopback HTTP is accepted. Tool and image support must match the task. The generic JSON provider supports explanations, not Copilot tool use.

Select the credential in AI settings. Saved sample sets are sent by built-in Copilots only when shareSampleValues is enabled for that credential; the default is off. This setting does not filter records returned by other MCP tools. Configure which operational records your external agent may pass to its model.

Explore AI-native workflows → · Read the API integration guide →

Give your agent a direct connection to the Center.