Install the solvethree agent skill
Download the official skill ZIP, inspect solvethree/SKILL.md, then copy the whole solvethree directory into your agent host’s skills directory. The skill supplies integration instructions—not an executable solver or an MCP connection.
On this page
Download, inspect and install
curl --fail --location --output solvethree-skill-0.1.0.zip \
https://solvethree.com/downloads/solvethree-skill-0.1.0.zip
unzip -l solvethree-skill-0.1.0.zip
unzip solvethree-skill-0.1.0.zip -d ./skill-review
# Inspect skill-review/solvethree/SKILL.md before installing it in your host.
# The archive contains instructions; it does not start an MCP server.- 1.
Read the extracted instructions
Open
skill-review/solvethree/SKILL.mdand read it before installing. This is the instruction text your agent will receive; downloading and extracting the archive doesn’t start a server. - 2.
Copy the directory into your host’s skills location
For Codex, use
~/.agents/skills/solvethree/SKILL.mdfor a user-level installation or.agents/skills/solvethree/SKILL.mdwithin a project. Copy the extractedsolvethreedirectory, not just the Markdown file. Other hosts have their own installation locations—don’t assume they use Codex’s paths. - 3.
Load it using your host’s documented process
Follow the host’s skill-loading instructions. Keep credentials in the host environment, never in
SKILL.md. A workspace API key must remain server-side.
The skill, MCP server and SDK do different jobs
| Component | What it provides |
|---|---|
| Agent skill | Instructions for integrating with the task API. Installing it doesn’t configure callable tools. |
| MCP server | A local stdio server exposing capabilities, task creation and task retrieval tools to a compatible host. |
| SDK | A library your application code uses to integrate with solvethree. |
If your host needs callable tools, configure the MCP integration separately. The official SDK download includes solvethree-mcp; its host environment supplies SOLVETHREE_API_KEY and SOLVETHREE_PROXY. It isn’t a hosted endpoint.
Don’t assume that loading the skill makes the agent call the API autonomously. It still needs a configured way to make requests and permission to use it. The MCP server doesn’t drive a browser or fill forms: your application must apply the returned solution.
Prepare the target before asking for a solve
For a reCAPTCHA task against a target you own or are authorized to test, provide the hostname, action and proxy input. The workspace needs an active site profile first; neither installing the skill nor sending a task creates one. Use Site profiles, or follow the profile setup instructions, and check that the profile matches the intended hostname, capture action and reCAPTCHA type.
Keep the task ID after creation. pending and solving mean keep waiting; only ready contains a usable solution. If the task is failed, stop polling and inspect errorCode and error rather than repeating unchanged inputs.
Apply the result in the request to its intended target using the returned browser identity and successful proxy. For reCAPTCHA, submit solution.token; other products return different result formats. A ready task means solvethree generated a solution—not that your application request succeeded.
Sources and API reference
Run your first task.
Start with $5 in workspace credit.