CorpAI Sandbox brings MCP evaluation to the customer's laptop

CorpAI Sandbox gives customers a fast way to experience the CorpAI workflow on their own machine. With Docker running and an LLM API key configured, users can open CorpAI in the browser, deploy MCP servers from the catalog, and ask the LLM to use those tools through natural language.

The goal is straightforward: make evaluation tangible. Teams can see how CorpAI connects language models to enterprise-style tools, how MCP servers are deployed, and how tool calls appear inside a governed chat experience before a larger rollout begins.

Why Sandbox Matters

Enterprise AI adoption often starts with a simple question: what would this feel like with our workflows? CorpAI Sandbox helps answer that question quickly. A team can run a contained sandbox, connect it to an LLM provider, and interact with real MCP tools from a familiar web UI.

This makes early evaluation more concrete for platform teams, security teams, and business stakeholders. Instead of reading about MCP servers in the abstract, users can deploy one, ask a question, watch the tool invocation happen, and inspect the resulting behavior.

How It Runs

CorpAI Sandbox is distributed as a release container image. Once started, it brings up the sandbox application stack and creates a Kubernetes sandbox for MCP server deployments. The browser UI runs locally, the API runs locally, and deployed MCP servers run as isolated workloads inside the sandbox.

Users bring their own LLM provider key. CorpAI Sandbox supports the same evaluation pattern across supported providers: configure the key, open the app, deploy a server from the catalog, and ask the assistant to use it. Tool calls are shown in the conversation so users can see when the LLM chose to invoke an MCP tool.

What Users Can Do

Users can browse the CorpAI MCP catalog, deploy selected servers into the sandbox, and connect those tools to chat. A simple example is deploying a time server and asking the assistant to check the current time in another city. CorpAI shows the answer and the tool invocation, making it clear how the assistant reached beyond the model response.

The same experience applies to more practical evaluation scenarios: credential-backed integrations, local infrastructure access, and workflow-specific tools. The sandbox gives teams a place to test the behavior of MCP-enabled AI interactions with a smaller operational footprint.

Who It Helps

Platform teams can use CorpAI Sandbox to understand the deployment model. Security teams can inspect how tools are exposed to the assistant and how executions appear in the product. Business teams can experience the user workflow without waiting for a full enterprise deployment.

The result is a shorter path from curiosity to hands-on evaluation. Customers can move from image pull to a running sandbox, from catalog selection to MCP deployment, and from a natural language prompt to a visible tool call in the same afternoon.