Add solvethree MCP to VS Code

Create or update .vscode/mcp.json with the configuration below. If the file already contains other servers or inputs, merge these entries rather than replacing them. VS Code must be able to run npx locally.

{
  "inputs": [
    {
      "type": "promptString",
      "id": "solvethree-key",
      "description": "solvethree workspace API key",
      "password": true
    },
    {
      "type": "promptString",
      "id": "solvethree-proxy",
      "description": "HTTP(S) target proxy URL",
      "password": true
    }
  ],
  "servers": {
    "solvethree": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "--yes",
        "--package=https://solvethree.com/downloads/solvethree-sdk-0.1.0.tgz",
        "solvethree-mcp"
      ],
      "env": {
        "SOLVETHREE_API_KEY": "${input:solvethree-key}",
        "SOLVETHREE_PROXY": "${input:solvethree-proxy}"
      }
    }
  }
}

The command runs solvethree-mcp from the official SDK download. The two password prompts supply SOLVETHREE_API_KEY and SOLVETHREE_PROXY to that process. Enter your workspace API key and HTTP(S) proxy URL in those prompts, not in chat or a committed configuration file. The proxy URL must not contain a path, query, or fragment.

The configuration shape is documented in the VS Code MCP configuration reference. This is a local process, not a hosted MCP endpoint.

Once VS Code has started the server, ask the agent to call solvethree_capabilities before requesting a solve. The server exposes these tools:

• solvethree_capabilities — inspect the service’s capabilities.
• solvethree_create_recaptcha — create a reCAPTCHA task.
• solvethree_create_kasada — create a standalone Kasada task; current support is limited to HTTPS targets on www.ticketmaster.com.
• solvethree_create_tmpt — create a Ticketmaster TMPT task for Event or Login.
• solvethree_get_task — retrieve a task’s status and result.

If those tools aren’t available, first check that the local process starts and receives both environment variables. Changing the target site or creating another task won’t fix a server-startup problem.

1. 1.

   Make the site profile active

   Set up the target in Site profiles with the correct hostname, capture action, and reCAPTCHA type. An active profile must be available to your workspace before solving. Supplying a hostname and action to the tool doesn’t create one. See site profile setup if this is a new target.

2. 2.

   Create one task

   Ask the agent to use solvethree_create_recaptcha with site as the hostname, the required action, and a unique requestId for this logical solve. The tool also accepts optional recaptchaType and waitMs; the type defaults to enterprise-v3. It reads its single proxy from SOLVETHREE_PROXY, so don’t add a proxy candidate list, type, or url to this tool call.

3. 3.

   Follow the task to a terminal status

   Each create call makes at most one task. If its status is pending or solving, use solvethree_get_task to retrieve it again. Only ready contains a usable solution. On failed, stop polling and inspect errorCode and error; failed solves are refunded. Don’t retry unchanged invalid inputs.

Keep the requestId if creation times out. It must contain 1–64 ASCII letters, numbers, underscores, or hyphens. Repeating the same task request with the same ID returns the existing task; changing the inputs while reusing that ID conflicts. Recover the existing task before asking for a new logical solve.

For reCAPTCHA, submit solution.token to the intended target—not the task ID. Use the returned userAgent and secChUa with the same proxy used for the solve. Switching IPs or mixing unrelated browser headers can break the submission.

The other tools return different material: Kasada returns a headers object to apply together, while TMPT returns the raw tmpt cookie value. Keep each result with its associated browser identity, proxy, and target.

The MCP host receives private solution data, but it doesn’t submit it for you. A ready task means the solution was generated, not that the website accepted your application request. If the target rejects it, inspect that outgoing request before paying for another solve.