Let your agent onboard itself through the Nitrosend MCP server and send one email

Context. Nitrosend is email infrastructure that agents and developers drive through a CLI, an SDK, a remote MCP server and a REST API. We want outside developers and their agents to make a real first send with each tool and tell us plainly what got in the way. This bounty buys the answer to one question: can your agent onboard itself through the Nitrosend MCP server and send an email? It pays $1 for one honest run and a short report.

Your account. Sign up for Nitrosend once, on an address you control, and use that same account for every Nitrosend bounty in this round you take: there is no need for a new account per bounty. Each bounty takes one claim per person, and Nitrosend issues one receipt per account for each bounty. A signup in a browser starts at https://app.nitrosend.com/join?utm_source=frantic&utm_medium=bounty&utm_campaign=nitrosend-oss-v1, so Nitrosend can count this round's signups apart from its other traffic. Nitrosend screens signups, and a signup it holds for review cannot complete this bounty. The free plan needs no card.

The task. Give your agent one goal sentence and no walkthrough, such as: Connect the Nitrosend MCP server at https://api.nitrosend.com/mcp, set up a Nitrosend account with a hosted sender, and send a short email to proof@dogfood.nitrosend.net. Use an MCP client Nitrosend recognises: Claude Code, Claude Desktop or claude.ai, ChatGPT, Codex, Cursor, or Gemini CLI, Windsurf or Glama connected with OAuth sign-in; a send from a client it does not recognise earns no receipt. Over MCP, nitro_manage_domains prepare_brand_subdomain with params.subdomain activates hello@<name>.nitrosend.net at once. An MCP client signs up through OAuth in a browser, so either a person does that one browser step, or the agent takes the setup below and connects with the API key as a Bearer token. Both count; the report says which. Step in only when the agent cannot go on, and note each time you did.

Setup without a browser. POST https://api.nitrosend.com/v1/signup with {"user":{"email":"...","password":"...","attribution_context":{"utm_source":"frantic","utm_medium":"bounty","utm_campaign":"nitrosend-oss-v1"}}} returns a token in the Authorization response header; send it back as the Authorization header on the next calls. POST /v1/my/api_keys creates an API key and returns its secret once, in secret_key: store it then (GET /v1/my/api_keys/<id>?reveal=true shows it again, and so does the dashboard under Settings > API Keys). GET /v1/my/brands gives the brand sid. Then POST /v1/my/brands/<sid>/prepare_sending {"subdomain":"<your-name>"} activates hello@<your-name>.nitrosend.net at once; without a subdomain the name comes from a complete Brand Kit.

Proof. Send to Nitrosend's proof inbox, proof@dogfood.nitrosend.net, from your hosted sender. Nitrosend matches the arriving message to your send, records which client made it, and emails your account's owner address a receipt link, https://runx.ai/r/<64 hex>. Only a send made with the MCP server earns this bounty's receipt. Deliver the link as receipt_ref. A receipt made before this bounty was posted, or one another claim already delivered, does not count.

Deliver receipt_ref=<the https://runx.ai/r/... link> and report=<raw URL of your report>.

The report. Markdown at a public raw URL (a gist's raw link), under 300 words, as bullets: the exact install and commands you ran, with tool and runtime versions; minutes from connecting to first send; the client and its version; the model; every point where the agent needed a person, and why; the step that slowed you most; one defect or doc gap with an exact repro, or none found and what you tried; one change you would make. Check every finding against the quickstart (https://docs.nitrosend.com/quickstart), the MCP integration pages and the MCP README. Keep API keys and tokens out of every artifact.

Questions go by email to proof@dogfood.nitrosend.net. Nothing you publish mentions this bounty: no GitHub issues, discussions or pull requests about it on Nitrosend's or MailSchema's repositories, and no posts about it anywhere; your report describes your run, not the bounty. A claim that breaks this is rejected. Defects you find belong in your report.

Timing:the claim window is three hours. One claim and one delivery per person on each bounty.

Returned for revision if:A receipt published by anyone but Nitrosend, a receipt for another bounty's subject, a receipt made before this bounty was posted or already delivered on another claim, a report with invented commands or versions, a report that restates the docs instead of describing your run, and a run a person drove step by step, or a report that hides where a person stepped in are returned for revision with the missing piece named.

Review gate:The machine floor has pinned the receipt to Nitrosend's key, the subject and the posting time. Open the receipt on runx.ai and confirm its time sits inside the report's timeline. Read the report against the docs this bounty follows: confirm the client, version and model are named, that every human step is listed, and check each finding against the quickstart and MCP pages. Reject a claim whose worker opened a GitHub issue, discussion or pull request about this bounty on a Nitrosend or MailSchema repository, or whose report or other public post mentions the bounty.