Automations and agents as tools

A profile can expose the automations and agents you have already built in Contentstack. Each Automate automation and each Agent OS agent appears as a single MCP tool alongside the Contentstack API tools. One tool call then runs the automation or agent, which performs the workflow steps, instead of the AI client calling each API operation in sequence.

Both appear to the client as tools. The table below sets out where they differ:

	Automation	Agent
How it runs	Executes the fixed steps you configured	Uses a language model to interpret the input, then runs actions
Input	Requires the inputs it declares	Accepts one free-form input object
Speed	Returns as soon as the steps finish	Adds model inference time to every call
Cost	No AI credits	Consumes AI credits on every call
Returns	The automation's response payload, or an acknowledgement	The agent's output

Prefer an automation when the work is deterministic. Use an agent when the request needs interpretation.

You select automations and agents in the same tool browser as the Contentstack API tools. Follow Step 1: Select tools, and in the browser open the Automations or AI agents tab instead of Contentstack APIs.

Both tabs list entries by project rather than by catalog. They appear only when your organization has automations or agents that qualify, as described in What qualifies. In an organization using neither, only the Contentstack APIs tools appear.

Contentstack lists an automation or an agent only when it can call it over HTTP. The table below lists what each one needs:

Requirement	Automation	Agent
HTTP trigger	Required as the first step	Required (HTTP Request Trigger)
Active	You activate it in Automate. Contentstack lists neither drafts nor paused automations	You activate it in Agent OS
Published	n/a	You publish it in Agent OS. Contentstack does not list a draft agent
Trigger ID	Required	n/a
Secure HTTP Trigger	Must be off. Contentstack cannot detect it on an automation, so a secured automation still appears and fails when the agent calls it	Must be off. Contentstack detects it on an agent, so a secured agent never appears

An automation or agent that meets every row in the table above is eligible, but eligibility does not guarantee it appears in the tool browser (see Limitations).

If your automation does not appear, activate it in Automate. If your agent does not appear, publish and activate it in Agent OS and confirm it has an HTTP Request Trigger. After each of these fixes, wait up to five minutes before reopening the tool browser. Contentstack caches the discovered list server-side, so if you reopen the tool browser sooner, it shows the same stale list.

Contentstack prefixes each tool name with its type and derives the rest of the name from the title you gave it. For example, the automation "Publish Weekly Digest" becomes automate_publish_weekly_digest. An agent becomes agent_<AGENT_ID>.

Contentstack builds the tool's input schema from the automation's own references. The two mappings below give the parameter kind each field becomes:

• Fields that the automation reads as {{1.body.*}} become body parameters.
• Fields that the automation reads as {{1.query.*}} become query parameters.

When an automation names its inputs, the tool's input schema lists each input as a parameter. An automation that reads a single unstructured payload produces a tool whose inputSchema declares no properties. Name the inputs in Automate before you expose the automation.

An agent takes one free-form input object. The agent's configuration defines the structure of this object. If you omit the object, the agent runs with its default instructions.

Contentstack uses the description you wrote in Automate or Agent OS. The model uses this description to select a tool. Write the description for the model, and state what the automation or agent does and when to call it.

Agent runs are asynchronous. The runtime polls until the agent finishes or the profile's timeout expires.

Warning An automation's HTTP trigger URL is unauthenticated. Anyone who has the trigger ID can run that automation without signing in, so treat a trigger ID as a secret and do not share it in tickets, chat, or source control.

Because of this, Contentstack excludes automations and agents from an exported profile. You re-add them manually after an import.

Profiles that added automations before Contentstack separated them into their own tab still run those automations, but the wizard cannot re-select those tools when you edit the profile. To edit such a profile, remove the automation and add it again from the Automations tab.

The Contentstack APIs tab includes a catalog named Automate. Its tools manage Automate itself, including listing and creating projects, inspecting automations, activating them, and triggering one by ID.

The Automate catalog and the Automations tab call different Automate endpoints with different credentials, so one cannot replace the other. The table below compares the two:

	Automate catalog	Automations tab
Calls	The Automate management API, with your Open Authorization (OAuth) token	The automation's own HTTP trigger URL
Covers	Any automation the API exposes to your account	Only automations that meet What qualifies
Returns	The management API's response	The automation's response payload, or an acknowledgement
Input schema	The management API's own parameters	Contentstack builds it from the automation's declared inputs

When to use each.

Use the Automations tab whenever the automation qualifies, and the Automate catalog only when you need to manage Automate itself.

Automate catalog tools need a project.

Select one in the profile's configuration, or have the agent supply project_uid on each call.

Additional Resource: For more information on automation and agent errors, for example a secure HTTP trigger error or a timeout, refer to the Profile Hub Automations and Agents section of the Contentstack MCP Troubleshooting Guides documentation.

• Discovery covers up to 50 projects per organization.
• Contentstack caches the discovered list for five minutes. A newly activated automation can take that long to appear.
• Discovery stops after eight seconds. If it does not finish, the tab lists fewer entries than exist.
• The profile's timeout caps an agent run at 120000 ms. The connector URL accepts a timeout value up to that maximum.
• You can still add a secured automation to a profile, and it fails when the agent calls it. See Troubleshooting.
• An exported profile contains no automations or agents. See Exports drop these tools.

• Create a profile: pick automations and agents in the tool browser's wizard step.
• Manage profiles: see what an export leaves out and the steps to re-add tools after an import.
• Monitor tool calls: read the Executions view to find the error message of a failed automation or agent call.