Connect AI Assistants with MCP

Giskard Hub includes a Model Context Protocol (MCP) ↗ server. With it, an AI assistant can work with the Hub on your behalf. You can ask it to list projects, register agents, create datasets, run evaluations and vulnerability scans, and inspect the results.

The server works with any MCP client that supports the HTTP transport, for example Claude Code, Codex, Cursor, or VS Code.

Before you start

You need two values:

• Hub Base URL: For the Giskard self-serve SaaS, this is https://saas.giskard.cloud/. If your organization uses an enterprise Hub, cloud or on-premise, use the URL of that instance.
• Hub API Key: Click the user badge in the bottom-left corner of the Hub UI and copy the API Key value. See Finding your API key.

Set them as environment variables:

export GISKARD_HUB_BASE_URL="https://saas.giskard.cloud/"
export GISKARD_HUB_API_KEY="<your-api-key>"

The MCP server endpoint is your base URL followed by /_api/mcp/. The server authenticates each request with the X-API-Key header. The assistant has the same permissions as the user who owns the API key. To control who can use API keys, see User Management.

Connect Claude Code

Run this command to add the Hub as an MCP server for your user:

claude mcp add --transport http --scope user giskard \
  "${GISKARD_HUB_BASE_URL%/}/_api/mcp/" \
  --header "X-API-Key: $GISKARD_HUB_API_KEY"

Then start Claude Code and run /mcp to make sure that the giskard server is connected.

Connect other MCP clients

Most clients use a JSON configuration file to declare MCP servers. Add an entry with the endpoint URL and the API key header. The exact file location and format depend on the client, so refer to its documentation.

{
  "mcpServers": {
    "giskard": {
      "url": "https://saas.giskard.cloud/_api/mcp/",
      "headers": {
        "X-API-Key": "<your-api-key>"
      }
    }
  }
}

Example prompts

After you connect, ask your assistant in plain language. For example:

List my Giskard Hub projects and show the latest evaluation results in the Support project.

Run a vulnerability scan on the customer-support agent and summarize the critical findings.