Developers

Everything Herd can do, you can call.

Herd, the REST API, the MCP server and the CLI all call the same tools. Connect Claude, Claude Code or Cursor with one URL, then ask who's waiting on a reply.

How it's built

One registry, four ways in.

Each tool is defined once. Herd, the API routes, the MCP tools and the CLI commands are all generated from it, so a new tool shows up everywhere at once.

One tool registryinput · output · scope · effect
Herd
REST API
MCP server
CLI

REST API

Safe by default.

Anything that sends, merges, archives, disconnects or removes access needs "confirm": true. Without it you get a 202 with a plain-English summary and a token to resend with.

Count interested replies still waiting
curl https://api.replysync.co/v1/threads \ -H "Authorization: Bearer rs_live_…" \ -G -d label=interested -d status=waiting
Base URLhttps://api.replysync.co/v1
AuthBearer key (rs_live_…), shown once, stored hashed
Versioningv1 only adds fields. Nothing is removed or renamed.
ErrorsWritten for people, with a request_id for support
SDKreplysync on npm, generated from OpenAPI
threads:readdrafts:writesends:writeclients:writeauto_mode:write
Only a key with sends:write can send. A key can never do more than the person who made it.

MCP server

Ask your AI who's waiting.

Hosted at mcp.replysync.co with OAuth. You only see your own clients, and anything that sends or deletes asks first. Comes with two prompts: triage_morning and weekly_client_summary.

CLI

Morning triage from your terminal.

Install with npm, log in once, and set a default client. Sending or deleting asks Proceed? [y/N] unless you pass --yes. --json returns the same shape as the API.

Webhooks

Know the moment it happens.

Every delivery is signed and carries an event ID for de-duplication. Failures retry 5 times over about an hour. After 20 failures in a row the endpoint pauses and we tell you.

reply.receivedreply.labeleddraft.approvedmessage.sentmeeting.bookedclient.createdauto_mode.report

Limits

Rate limits.

Sends are queued under your sequencer's own limits and never rejected. Limits may change before launch.

TypeLimit
Reads600 per minute per key, up to 20 at once
Writes120 per minute per key
HeadersRateLimit-* on every response, Retry-After on a 429
AuditEvery API, MCP and CLI call is logged: who, what and when
Hehe, that tickles!

Developers

Get a key
and build.

Access opens in small groups, so create an account first. Your key lives in the app.

Say hi to Herd and get set up

Already have an account? Sign in