OpenClaw provider plugin for GrowthCircle.id models
翻訳待ち
プラグインTypeScript · PythonGitHub @Growth-Circle自分でインストール所有者未確認
Paste into Claude or ChatGPT for installation instructions suited to your environment.
Have you tried it? One tap is enough. No sign-in needed.
The maker's introduction consists of the line above. Read the repository README for more details. Repository README ↗
Loading the full README.
Read the full README ↗GrowthCircle.id for your AI coding tools — one command, no manual config editing.
export GROWTHCIRCLE_API_KEY="<your-growthcircle-key>"
npx gc-provider setup
That detects the AI coding clients installed on your machine, shows you exactly what it will change, and writes the provider config after you confirm.
ai.growthcircle.id serves three wire formats, which is why so many different
tools can talk to it directly with no proxy in between:
| Wire format | Endpoint | Used by |
|---|---|---|
| OpenAI Chat Completions | https://ai.growthcircle.id/v1 | OpenClaw, Zed, opencode, Kilo, Cline, Roo, Cursor, Trae Agent, Hermes |
| OpenAI Responses | https://ai.growthcircle.id/v1 | Codex CLI |
| Anthropic Messages | https://ai.growthcircle.id/anthropic | Claude Code |
Model discovery is always GET https://ai.growthcircle.id/v1/models.
Checked against a live paid key on 2026-08-07:
| Path | Tool calling | Status |
|---|---|---|
/v1/chat/completions | yes | Working — returns proper tool_calls |
/v1/responses | no tools | Working |
/v1/responses | with tools | Broken upstream — see the Codex note below |
/anthropic/v1/messages | — | Routes correctly; blocked by upstream Anthropic credit |
Every client except Codex CLI uses /v1/chat/completions, which is fully
working.
| Client | How | Status |
|---|---|---|
| Codex CLI | npx gc-provider setup codex | Automated — blocked by a server-side bug, see below |
| Claude Code | npx gc-provider setup claude-code | Automated |
| Zed | npx gc-provider setup zed | Automated (key pasted once in Zed's UI) |
| opencode | npx gc-provider setup opencode | Automated |
| Kilo Code | npx gc-provider setup kilo | Automated |
| Trae Agent (CLI) | npx gc-provider setup trae-agent | Automated (per project) |
| OpenClaw | openclaw plugins install clawhub:gc-provider | Native plugin |
| Hermes Agent | npx gc-provider gc-provider-install-hermes | Native plugin |
| Cline | Settings panel | Manual — no config file to write |
| Roo Code | Settings panel | Manual — no config file to write |
| Cursor | Settings → Models | Manual, with caveats |
| Windsurf | — | Not supported by the vendor |
| Trae IDE | — | Not supported (Trae-AI/Trae#2076) |
| Antigravity | — | Not supported by the vendor |
| Kiro | — | Not supported (kirodotdev/Kiro#9367) |
"Manual" and "Not supported" are limits of those products, not of GrowthCircle. Neither Cline, Roo, Cursor, Windsurf, Trae IDE, Antigravity nor Kiro exposes a provider config file that a CLI could safely write.
There is no MCP server in this package, and there will not be one: MCP adds tools and resources to a host, it cannot change which model a host talks to.
20 or newer for the setup CLI (22.19.0+ for the OpenClaw plugin on
the current latest release line).gc-free, gc-paid,
gc-team.To create a key:
Then export it, and add the line to ~/.bashrc or ~/.zshrc so it survives a
new shell:
export GROWTHCIRCLE_API_KEY="<your-growthcircle-key>"
Verify the key works before configuring anything:
curl https://ai.growthcircle.id/v1/models \
-H "Authorization: Bearer $GROWTHCIRCLE_API_KEY"
The key tier decides the model list. Free keys use -free model ids, for
example gpt-5.6-luna-free.
npx gc-provider setup [client...] # write provider config (default)
npx gc-provider uninstall [client...] # remove exactly what setup wrote
npx gc-provider status # what is detected and configured
Client names: codex, claude-code, zed, opencode, kilo, trae-agent.
With no name, every detected client is targeted.
| Option | Effect |
|---|---|
-y, --yes | Apply without the confirmation prompt |
-n, --dry-run | Print the diff and exit without writing |
--all | Target every client, not only the detected ones |
--offline | Skip the live /models lookup, use the bundled catalog |
--tier free|paid|team | Force a tier instead of reading the key prefix |
-h, --help | Usage |
-V, --version | Version |
Your API key is never written to a config file. Every generated config
references the environment variable by name (env_key, {env:...}), so
rotating a key means changing one export, and a leaked config file leaks
nothing.
Existing config is merged, not replaced. Only the GrowthCircle block is
inserted or updated. Comments, key order, and indentation width in your files
are preserved — including the comments in a hand-tuned Zed settings.json.
Every modified file is backed up first to <file>.bak.<timestamp>.
Re-running changes nothing. A second setup with the same catalog reports
Nothing to do.
Unparseable files are refused, not overwritten. If a config file is not valid JSON or YAML, that client is skipped with a message.
The network is never required. With a key set, the live /models list is
used. Without one, or if the lookup fails, the bundled tier list is used and the
CLI says which it chose.
This is the beginning of the README. Read the rest at the original. Read the full README ↗