CLI Tools

Codex CLI

Add a provider in your user configuration and use the Responses API.

ConfigurableOpenAI ResponsesDocs checked 2026-09-18

The vendor documents custom endpoints. Check model capabilities and verify the connection.

On this page

Before you begin

Install Codex CLI and choose a model and key group that support Responses and tool calling. Follow the official Windows guide for native or WSL setup.

Have an A2Agent API key ready and copy an accessible model ID from the catalog. Replace every placeholder, back up existing settings and keep keys out of repositories.

Manage API keys Browse models

Setup

File contents are usually shared; switching OS changes commands and paths. For WSL on Windows, select macOS / Linux.

01Install Codex
With Node.js installed, run the command below, or use a binary from the official installation page.
Open official documentation
Shell
```
npm install -g @openai/codex
codex --version
```

02Add a provider

Back up your user config.toml. Merge these fields while preserving existing settings; put top-level fields before TOML tables.

~/.codex/config.toml

model = "YOUR_MODEL_ID"
model_provider = "a2agent"

[model_providers.a2agent]
name = "A2Agent"
base_url = "https://a2agent.me/v1"
env_key = "A2AGENT_API_KEY"
wire_api = "responses"

03Set the key and launch
Pass the key through the environment; keep it out of the repository.
Shell
```
export A2AGENT_API_KEY='YOUR_A2AGENT_API_KEY'
codex
```

Capabilities and limits

Chat Completions cannot replace Responses here. Do not append /responses to base_url.
The Codex editor extension may read the same user configuration. Record the previous provider and model before editing.

Verify the connection

Replace every placeholder and run locally. Expect JSON containing OK. Then start a new client conversation, test one read-only tool call separately, and confirm the requests in A2Agent usage records. This test incurs a small amount of model usage.

Shell

curl --fail-with-body --silent --show-error --max-time 60 'https://a2agent.me/v1/responses' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer YOUR_A2AGENT_API_KEY' \
  -d '{"model":"YOUR_MODEL_ID","input":"Reply with OK."}'

Troubleshooting

If the environment variable is missing, launch from the terminal where you set it. Check env_key and the TOML structure.

401 / 403: authentication or access denied

Use a key issued by A2Agent, without extra spaces, and check that it is active and unexpired. Confirm the group permits the model and protocol. A vendor subscription credential cannot replace an A2Agent key.

404 / model not found: endpoint or model mismatch

Check the root URL, /v1 prefix and full request path against this guide; avoid duplicated paths. Model IDs are case-sensitive and must be accessible to the current key.

429: quota or rate limit

Check your balance, key limits, concurrent usage and service status. Wait before retrying and reduce simultaneous agents; repeated retries do not remove a quota limit.

Chat works, but tools fail or streaming stops

Test a short text request, then a single read-file tool call. Check tool support, output budget, proxy timeouts and client version. Inspect the diff before resuming a long task to avoid repeating actions.

Restore previous settings

Restore the previous model and model_provider, remove the a2agent table, and close the terminal containing the key.

Sources and check date

Status is based on official configuration docs, not end-to-end client testing with A2Agent.

Docs checked 2026-09-18