Skip to main content
By the end of this guide, your coding agent can identify your Apostra account and read its first objects. You need an Apostra account and an MCP client that supports remote Streamable HTTP. Account creation, missing permissions, or seller onboarding may take longer than this connection step.

1. Connect MCP directly

Add Apostra’s remote Streamable HTTP MCP connection to your coding agent, then complete browser sign-in and consent. The connection uses your existing account permissions. Apostra’s API and sign-in URLs currently use interchange.io.
Complete OAuth in the browser, then open a Codex session in your project. See Codex MCP setup for client configuration and OAuth controls.
Use an existing connection if it already supplies this endpoint. The coding agent should not ask you to paste credentials into chat.
The native apostra-developer plugin, Build with Apostra skill, and TypeScript and Python starters are pending publication and fresh-client verification. Installation instructions and starter links will be added after those checks pass. Use the direct MCP connection above to get started today.
For a headless application, use an appropriately scoped API credential from a secret manager. Do not put it in a prompt or committed client configuration. See Authentication for supported credential types.

2. Prove account access

Ask your agent:
If your client reports that the connection needs authentication, open its MCP controls and complete OAuth before retrying. Adding a server is not by itself proof that sign-in has completed. get_status is a read-only check. A successful result proves that the connection can reach the account. It does not prove that every advertiser, seller, or write operation is authorized. If the wrong account is selected, call switch_account with an accountId returned by the status result, then call get_status again. An Organisation or unclassified account may offer navigation without buying or selling tools. See Account access.

3. Make a first read

Ask the agent to inspect tools/list, then call search with the input that matches your active account:
This lists seller targets and their buyer-specific connection and readiness information. It does not contact sellers with a campaign brief. An empty result is a valid read, not proof that the connection failed.
You now have a verified MCP connection and a first account-scoped read. The schema returned by tools/list is the contract for your session; tools can differ by permissions and feature availability.

4. Build a small prototype

Start with a read-only seller-discovery prototype in a Buyer Account:
Keep the two checks separate: a fixture test proves local code handles the expected responses; the live get_status and search calls prove account access. Neither proves that a campaign can launch. Continue with Build an agent to add durable buying steps, or Build a reporting pipeline to read existing data.

If the first call fails