Building with AI agents usually hits a wall at the integration layer. The agent exists, your data platform exists, but connecting them requires writing code to translate between the two. Each new capability means more integration work. The automation potential is there, but unlocking it requires development effort that shouldn't be necessary.
And the need for this automation is only growing. Using agents to speed up AI data workflows is becoming essential for scaling annotation tasks, whether that's pre-annotation, LLM-as-a-Judge workflows, or managing project progress at scale. But agents can't reach their full potential if every interaction with your data requires custom scripts and intermediate actions.
Today, we’re excited to lift that burden with the SuperAnnotate MCP Server, to let you connect agents directly to SuperAnnotate projects without writing custom code.
What is the SuperAnnotate MCP Server?
The SuperAnnotate MCP Server is a tool that lets you create AI agents in your pipeline that interact directly with SuperAnnotate projects, folders, and items without the need for any intermediate custom action nodes. Before MCP, building these workflows usually meant adding custom action nodes and writing extra Python just to fetch project data, inspect annotations, and push updates back into SuperAnnotate. Now agents can do all of that directly through MCP tools with a simple prompt, no intermediate glue code required.
This essentially turns your agent into a first-class participant in the SuperAnnotate platform. Your models can leverage SuperAnnotate's core project management capabilities just like a human user would.
Under the hood, the server uses Model Context Protocol (MCP), a lightweight approach to expose actionable endpoints that agents can use to:
- list and filter projects, folders, and items
- retrieve project, folder, and item metadata
- read and update item annotations
- change item status
- query workflow statuses
All of these operations become accessible as callable tools inside your agent configuration.
How SuperAnnotate MCP Server Works
The setup happens entirely in Agent Hub, which brings your models into annotation workflows without needing custom integrations. The MCP Server opens the door to agents that can interact with real project data directly, executing the tasks you define.
Four steps get you there.
Step 1: Connect a Model in Agent Hub
To get started, head to Agent Hub and click + Connect Model. Here, you’ll choose your provider (such as OpenAI, Anthropic, and others), select the model, and configure credentials along with advanced settings like Temperature.
Step 2: Add the SuperAnnotate MCP Server
During model configuration, under Advanced Settings, you’ll find a new SuperAnnotate MCP Server option in the Tools section for supported providers like OpenAI and Anthropic.
Step 3: Configure the MCP Tools
Once you open the MCP Server configuration view, you’ll see a list of all supported MCP tools.
- Each tool has a description and defined parameters (like project_id, folder_id, item_id, etc.).
- Tools are checked by default – uncheck ones you don’t want your agent to use.
- Then simply click save to create your model with the MCP tool.
This gives you fine-grained control over what your agent can do, preventing risky or unintended actions. Note that you must keep at least one tool selected or you’ll get an error.
Step 4: Save and Use in Your Pipeline
After saving, you’ll see a SuperAnnotate MCP chip added to your model’s tool list in Orchestrate pipelines.
To set up your pipeline, connect the agent node to an event trigger node and write a detailed prompt with instructions for what you want the agent to do. The event trigger payload automatically injects the relevant project, folder, item, and component IDs, so you don't need to manually specify those in your prompt. Save the pipeline and it's ready to run.
Step 5: Review Agent Traces in Monitoring
Once your agent with the SuperAnnotate MCP Server has run, you can see exactly what it did through the Monitoring → Agents → Traces view. This trace view lets you review every MCP tool call your agent made, giving you observability into its behavior and helping with debugging, auditing, and optimization.
In the Traces tab in the pipeline’s monitoring section:
- Each agent invocation shows a list of tool calls
- Each tool block is collapsed by default – click to expand and view the Request and Response examples.
Why Agent Integration Matters
The SuperAnnotate MCP server helps anywhere you'd want an agent to autonomously manage or evaluate data within your projects.
Before this feature, connecting agents to SuperAnnotate typically required custom actions or extra configuration. Now:
- You configure tools with one UI, no hand-rolled code needed.
- Your agent can interact with real project data – listing items, pulling annotations, updating statuses, and more.
- You maintain control over what actions the agent can perform by selecting only the tools you trust.
Looking Ahead
With the SuperAnnotate MCP Server, we’re empowering teams to design more intelligent, responsive, and automated data workflows – keeping human expertise where it matters and letting smart agents handle the rest. Whether you’re building annotation automation, quality checks, or hybrid evaluation systems, this capability brings your agent’s power closer to your data.
Add the SuperAnnotate MCP Server to your model and see how agents can level up your workflows without extra custom actions.
