PTC MCP + CLI — translate your app with your coding agent

PTC is your expert AI translator. It learns your app, researches unclear texts, translates in your own style and tests the results.

Your agent needs both: the PTC MCP and this CLI (the agent runs it to scan your repo and report what it finds). There is no MCP-only path. Tested with Claude Code; any agent that speaks MCP and can run a shell command should work.

Connect your agent

1. Get your agent token

Sign up to PTC and open PTC's agent token page (https://app.ptc.wpml.org/dashboard/agent-tokens). Your agent token is yours (one per user): PTC mints it the first time you need it, and you can only rotate it, never add a second one. It is how your agent identifies itself to PTC.

2. Add PTC to your agent

The MCP endpoint is https://app.ptc.wpml.org/mcp.

Copy the setup for your agent from the agent token page: it has a ready block for Claude Code, Cursor, VS Code and any other MCP client, with your token already filled in. Paste it as it is.

Codex — first export PTC_API_TOKEN in your shell with the value the agent token page shows you (Codex reads the token from that variable), then:

codex mcp add ptc --url https://app.ptc.wpml.org/mcp --bearer-token-env-var PTC_API_TOKEN

3. Install the CLI

From the repository root:

mkdir -p .ptc/bin && curl -fsSL https://app.ptc.wpml.org/cli/ptc-cli.sh -o .ptc/bin/ptc-cli.sh && chmod +x .ptc/bin/ptc-cli.sh

Every ptc command in these docs is .ptc/bin/ptc-cli.sh run from the repository root (alias it as ptc if you like). .ptc/ is the CLI's own git-ignored folder, so later syncs reuse the script with no reinstall.

Needs bash and curl; the agent commands also need python3 (>= 3.6).

Then save your agent token for the CLI yourself (your agent asks you to; the token never goes through it):

.ptc/bin/ptc-cli.sh auth login

It signs you in through the browser: it prints a link and a code (and opens the link), you approve the code on PTC, signed in as yourself, and the CLI receives your agent token; you type no token and nothing is shown. On a machine with no browser, open the link on any device. --paste pastes the token from the agent token page instead (nothing is shown as you paste), and a token piped on stdin is read as before. It checks the token with PTC and saves it, readable by you only (mode 600), to %APPDATA%\ptc\credentials on Windows (Git Bash, MSYS, Cygwin) or ~/.config/ptc/credentials elsewhere ($XDG_CONFIG_HOME/ptc/credentials when set). A token PTC refuses is not saved. PTC_API_TOKEN or PTC_AGENT_TOKEN in the environment, or --api-token, still win over the file.

4. Give your agent one prompt

Localize this app into German and Spanish with PTC.

That is the whole instruction. PTC guides your agent through the rest.

What happens next

1. Your agent scans the repository (ptc scan). PTC learns the app, its features and its terminology.
2. PTC asks you only what the code can't answer. The questions arrive in your agent's chat; decisions such as which languages to translate into stop the run until you answer. PTC shows the price and waits for a yes before any paid work; a trial continues within its free cap.
3. PTC translates with your terminology and reviews every language in your app's context.
4. Before delivering, PTC asks whether you want the translations as a merge request (PTC set up in your CI) or via the API. Answer merge request and PTC hands your agent the CI file with that answer; the agent adds it to your repository (see CI). Answer API and your agent delivers with ptc sync.

Go deeper

• Agent commands: how PTC drives your agent, and the CLI commands it runs along the way.
• CLI reference: usage, options, configuration file, workflow, exit codes, troubleshooting.
• CI: pinning the CLI, GitHub Actions, GitLab CI.

License

MIT License