Run Go-to-Market from Codex, Cursor or OpenCode

Codex, Cursor and OpenCode run go-to-market work with Cargo the same way: install the cargo-ai CLI, sign in with cargo-ai login --email, install the Cargo skills for your agent, and ask for the job. Only three things change from one agent to the next: the flag on the skills installer, whether a plugin is available, and where the MCP server is configured. This page gives one setup and a table for the differences.

The shared part #

npm install -g @cargo-ai/cli
cargo-ai login --email you@company.com                 # sends a code, then exits
cargo-ai login --email you@company.com --code 123456
cargo-ai whoami                                        # must name a user and a workspace

Sign-in runs in the terminal, with no browser, and creates the account on first use. A new account starts with 100 free credits and no card. The CLI recognises when Codex, Cursor or OpenCode launched it and never opens an interactive prompt there, so an agent session cannot hang on a picker.

Or paste one line into the agent and let it do all of this:

Install Cargo by following https://api.getcargo.io/INSTALL.md

What differs per agent #

The skill files themselves are identical everywhere: plain SKILL.md folders, the same ones Claude Code uses.

Install the skills #

One command, with your agent’s flag:

npx skills add getcargohq/cargo-skills --agent codex -g -y      # Codex
npx skills add getcargohq/cargo-skills --agent cursor -g -y     # Cursor
npx skills add getcargohq/cargo-skills --agent opencode -g -y   # OpenCode

Drop -g to install into the current project (.agents/skills) instead of your home directory. For one job rather than the whole platform, install a job-named skill such as find-work-email: npx skills add getcargohq/gtm-skills/find-work-email.

The plugin, on Codex and Cursor. Both agents can also install Cargo as a plugin, which carries the same skills plus an approval hook: read-only cargo-ai calls run without a prompt, while sign-in, token minting, deploys and deletions still ask. On Codex, run codex plugin marketplace add getcargohq/cargo-skills, then install “cargo” from the Plugins menu. On Cursor, open Customize in the sidebar, add the getcargohq/cargo-skills marketplace and install Cargo. Use the plugin or skills add, not both, or every skill registers twice. OpenCode has no Cargo plugin, so the skills are the whole install there.

Connect the MCP server (optional) #

The skills drive the CLI, which covers everything. The MCP server lets the agent search, price and run actions and read data models as tool calls. cargo-ai mcp serves it over stdio with the credentials you signed in with, so no token goes in the config file.

Codex:

codex mcp add cargo -- cargo-ai mcp

which writes this to ~/.codex/config.toml:

[mcp_servers.cargo]
command = "cargo-ai"
args = ["mcp"]

Cursor, in ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):

{
  "mcpServers": {
    "cargo": {
      "command": "cargo-ai",
      "args": ["mcp"]
    }
  }
}

Cursor can also connect to the hosted server over HTTP and sign in with OAuth: use "url": "https://mcp.getcargo.io/mcp" in place of command and args.

OpenCode, in opencode.json at the project root or ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "cargo": {
      "type": "local",
      "command": ["cargo-ai", "mcp"],
      "enabled": true
    }
  }
}

The tools the server exposes are listed in the MCP server docs; the Cargo MCP server page covers the hosted endpoint and curated servers.

Run a first job #

Ask in plain words, for example “Find the work email for Jane Doe at acme.com”. The agent loads the work-email skill and runs prospeo.findEmail through the CLI, at 0.5 credits per record, returning the address or an empty value when nothing resolves. Across Cargo, work-email lookups ran 1,850,404 times and 63.6% of runs returned an address (Cargo platform data, 90 days to Oct 16 2026), so sample 10 to 20 records before a full list.

The full walkthrough of that job, with the exact command and what comes back, is in run go-to-market from Claude Code; every step after sign-in is the same in these three agents. When the job should run on its own, have the agent declare it as code with cargo-ai project init, as described in run go-to-market with coding agents.

Frequently asked questions #