The safety model is what makes this worth handing to an agent: writing is off until the server is started for it, every scenario tool dry-runs by default, and the dangerous ones carry their own ceilings — a maximum number of deletions, a change limit on batch apply, a separate confirmation before replacing redirects or adding a second SPF record. That turns the risky part of domain work into something reviewable. Start with the health check on a domain you already run to see what it reports before letting it change anything.
A Porkbun client in two layers: direct API calls for domains, DNS records, DNSSEC, SSL and URL forwarding, and scenario tools that fold a whole operator task — a cutover, a bulk change, an email setup — into one planned operation.
- Domains listed, checked for availability singly or in bulk, registered, auto-renew set, nameservers and glue records read and changed
- DNS records listed, read by id or by type and subdomain, created, edited and deleted, with DNSSEC records managed alongside
- URL forwarding rules read, added and removed, and the SSL certificate bundle retrieved
- A unified record lookup that accepts either a record id or a type and subdomain, returning matches, count and which selector was used
- Upsert: a record brought to the state you describe — created if missing, edited if present, no-op if already right — reporting before, after and what changed, and refusing when several records match ambiguously
- Deletion with a maximum, which hard-fails rather than proceeding when the plan would remove more records than you allowed
- A batch apply for migrations: plan first and read the create, edit, delete and no-op diff, then apply with explicit confirmation and a ceiling on how many changes are acceptable
- A guided web cutover producing steps, planned changes, verification and warnings — including that propagation timing is not something the tool can promise
- Redirects converged to a desired set, with merge as the default and replace requiring its own confirmation so a mass cleanup cannot happen by accident
- Beginner-safe helpers: apex and www set up in one call, a host's A and AAAA pointed at a new server IP, one subdomain record created or edited, and baseline email DNS for Google Workspace, Protonmail or your own records — which asks for confirmation rather than adding a second SPF record
- Read-only diagnostics: a health check across nameservers, DNS, DNSSEC, SSL and forwards with an overall status and recommendations, and a beginner-facing audit that warns about the apex, www and email basics
Porkbun API credentials — PORKBUN_API_KEY and PORKBUN_SECRET_KEY, both from the API Access page in your account. Account-level keys are not enough on their own: each domain has to be marked as available via API in its own settings, or calls against it fail. Node.js 20 or newer; it runs as npx -y porkbun-mcp. The server starts read-only, and writes stay refused until it is launched in write mode with the --get-muddy flag or PORKBUN_GET_MUDDY=true. Scenario tools default to dry_run: true, so a changing call reports its plan until you ask for the change. Bulk availability checks are capped at 100 domains and run sequentially by default with an 11-second delay, to stay inside Porkbun's rate limits.
One command plus a key — npx -y porkbun-mcp, then supply credentials