MCP

Tarsana has a hosted MCP server, so an AI assistant or agent can use your Tarsana account directly. It offers one tool for every API operation, with the same name, the same arguments and the same answers.

Connect

Create the access token with issue-access-token, as in step 5 of Build your first image. Give each client its own token, with a token_label that says which client it is, so that you can revoke one without the others.

Most MCP clients take a configuration like this one:

{
  "mcpServers": {
    "tarsana": {
      "type": "sse",
      "url": "https://mcp.tarsana.io/sse",
      "headers": {
        "X-Tarsana-Access-Token": "YOUR-ACCESS-TOKEN",
        "X-Tarsana-Tenant": "acme"
      }
    }
  }
}

Every request is checked against your access token before it reaches a tool. A request without a valid token is refused, and you get no tool list.

Call a tool

A tool takes the operation's parameters as arguments. A request body, such as a spec, is the body argument. For example:

{
  "method": "tools/call",
  "params": {
    "name": "status",
    "arguments": {
      "tenant": "acme",
      "submission": "sha256:cdff062988d4722b2ff052c97ebefb8b571858c493ce47b159df73b1c88809b1"
    }
  }
}

The answer is the same JSON the API returns. A refused call comes back as a tool result with isError set, and its text is the API's refusal, with the same error object, code and doc_url; see Errors.

Tools that only read have the read-only hint set, so your client can run them without asking you first. Tools that change something, and especially the ones marked Destructive, should be confirmed by you.

Tools

View this page as Markdown