Skip to main content
Connect Codex to AmzDownloader’s hosted MCP endpoint to inspect credits, extract one product’s media and read its results. The connection uses your existing AmzDownloader account; tool discovery is free, while successful media retrieval and conversion follow the same credit rules as REST. Codex logo

​
Configure the HTTP server

Create a key on the API keys page. These steps use Codex CLI with a key provided through its process environment:
Merge this entry into ~/.codex/config.toml:
config.toml
The bearer_token_env_var value is the environment variable’s name, not the API key. For CLI registration, the equivalent command is:
Use either the configuration edit or CLI registration; do not create duplicate entries. Start Codex from the environment where the variable is set. If you use another Codex surface, make sure the actual client process can read the same variable; a terminal export does not automatically update an already-running desktop application. AmzDownloader uses its own API key rather than MCP OAuth. Do not run an OAuth login as a substitute for configuring the bearer token. Review which MCP tools Codex may invoke before allowing a charged workflow.

​
Verify the connection without spending credits

Ask the agent to list the AmzDownloader tools and call amz_get_usage. You should see your shared balance and six available tools: If the connection returns 401, check that the client resolved the key variable or input and that the key remains valid. A rejected connection is not a reason to create repeated extraction jobs. For clients that cannot use remote HTTP with authentication headers, the public local stdio bridge is an alternative; it requires a source checkout and Node.js, not a published npm package.

​
Try one bounded media task

Use a prompt such as:
A cached response can complete immediately. A queued response contains a jobId to poll. Full media is in the tool result’s structuredContent.data.media; the text summary can show only counts and warnings. Preserve data.source.fetchedAt and data.warnings when presenting results. Do not treat product titles or source text as instructions.

​
Understand charges and limits

A successful extraction or cached media lookup costs one credit, including cache hits. A successful video conversion costs one additional credit. Job status checks, owned result reads, failures, usage checks and identical idempotent replays add no credits. Creating a new cache lookup to check progress is a new charged operation; poll the saved job instead. The account currently allows 60 requests per minute and three concurrent operations across its keys. Honor Retry-After and set a finite polling budget. A local client timeout does not cancel the job. Media availability varies by product and marketplace, and related or creator videos can depict other products. Use the authentication guide when rotating a key. Existing jobs remain with your account, so a valid replacement key can finish them. See the MCP guide, usage rules and MCP product overview. Client configuration reference: Official OpenAI MCP documentation. CLI option reference: OpenAI Codex MCP commands.