Sign inSign up free
Omnichannel inbox Visual job flows Characters Any model + fallbacks CRM actions & booking Knowledge bases Approvals & autonomy Test & Debug A/B testing Analytics Website chat widget Quote desk Custom tools MCP server Rate limits & queue AI appointment setter All features Pricing FAQSign up freeSign in
MCP server

Build and run your agents from Claude Code, Cursor, anywhere

Everything in ReplySetter can be driven over the Model Context Protocol. Create a token, add one line to your MCP client, and ask your coding assistant to build an agent, test it, read the trace and apply fixes, with the same validation as the UI.

Try it free How it works
In a terminal, the ReplySetter MCP server is added to Claude Code and asked to build and test a booking agent; tool calls stream and the agent appears in the app~/clinic · zsh$claude mcp add --transport http \ replysetter https://app…/api/mcp \ --header "Authorization: Bearer ia_…"Added MCP server replysetter>Build a booking agent for BrightSmile Dental and test it with aprice objection.replysetter.create_characterreplysetter.create_agentreplysetter.validate_flowreplysetter.create_test_sessionreplysetter.send_test_messagereplysetter.diagnose_test_turn● 6 tools called · 14.2 sReplySetter · AgentsBright Smile BookingCharacter: Mia · created via MCPFLOWStartQualifyBookStopflow validTEST · Price objectionIsn't $199 a lot fora whitening session?Fair question! It includes thetake-home kit, which is $80 onits own. Want Thursday at 10?diagnose: no issues found
130+MCP tools
1command to connect
Samevalidation as the UI
Scopedpersonal API tokens
Deep dive

A closer look at MCP server

01 · Tools

Agents, flows, tests, approvals, quotes and more

Create a personal API token, add the server to your MCP client with one command, and every part of the app is a tool call away: agents and characters, flow validation, test sessions with diagnose and re-run, approvals, knowledge bases, widgets, quote desks, suppliers, buyers and the queue.

  • Build, validate and simulate flows
  • Run, diagnose and re-run test conversations
  • Approve or reject actions and reply to threads
  • api_request reaches any other API route
A grid of MCP tool names lights up group by group: agents, testing, operations, knowledge and widgets, and the quote deskAGENTS & FLOWScreate_agentupdate_agentvalidate_flowsimulate_flow_turnlist_flow_templatesTESTINGcreate_test_sessionsend_test_messagediagnose_test_turnrerun_test_turnrate_test_turnOPERATIONSlist_actionsapprove_actionreply_to_threadget_queue_summaryretry_queue_jobsKNOWLEDGE & WIDGETScreate_knowledge_baseadd_documentsearch_knowledge_basecreate_widgetlist_widget_sessionsQUOTE DESKcreate_quote_requestrun_quote_stepsend_supplier_emailslist_supplierssend_offer+ 100 more, and api_request for any other route
02 · Tokens

Personal tokens with a tight scope

Tokens act as you, so they can do what you can in the app. But they only authenticate API calls: they can't mint or revoke tokens, and they can't reach auth or OAuth routes, so a leaked token can't lock you out or take over connections.

  • Bearer personal API tokens
  • Can't mint or revoke tokens
  • No access to auth or OAuth routes
  • Revoke any token from Settings
A personal API token can reach agents, threads and actions but is refused at the token, auth and OAuth routesAPI tokenia_7f2c…e91acts as you/api/agentsbuild & edit/api/threadsread & reply/api/actionsapprove/api/tokenscan't mint tokens/api/authno auth routes/api/oauthno OAuth
03 · Same rules

Same validation as the UI, by design

Each tool forwards to the matching REST route as the token's user, so validation, permissions and side effects are identical to clicking the button in the app. An invalid flow is refused with the same message either way.

  • Every call runs as the token's user
  • Identical validation and side effects
  • Streamable HTTP, stateless: works behind any load balancer
  • Tools that send real messages are clearly marked
Publishing from the UI and calling update_agent over MCP both go through the same REST route and validation, returning the same errorApp UIPublish buttonMCP clientupdate_agentPUT/api/agents/:idSame validationsame side effects,same permissionsStep “Book” has no exitsshown in the builderStep “Book” has no exitsreturned to your MCP client
Under the hood

The details

Transport

Streamable HTTP at /api/mcp, stateless.

Auth

Bearer personal API token.

Agents

create_agent, update_agent, validate_flow, simulate_flow_turn.

Testing

create_test_session, send_test_message, diagnose_test_turn, rerun_test_turn.

Operations

list_actions, approve_action, get_queue_summary, retry_queue_jobs.

Escape hatch

api_request reaches any other /api route.

Use cases

Who it's for

Agent engineering

“Build a booking agent for a dental clinic and test five objections.”

Ops from the terminal

“What's failing in the queue? Retry the rate-limited jobs.”

Bulk setup

Spin up characters, knowledge bases and widgets for a new client in one prompt.

FAQ

MCP server: common questions

Which MCP clients work?

Any client supporting Streamable HTTP, including Claude Code and Cursor.

Can a token do everything I can?

It can call the API as you, but it can't mint or revoke tokens or reach auth and OAuth routes.

Can MCP send real messages?

Yes, a few tools do (reply, approve, send quote drafts); they're clearly marked.

Is behaviour different from the UI?

No. Each tool forwards to the same REST route with the same validation.

Keep exploring

Works hand in hand with

Developers

Custom tools

Connect any HTTP API as a tool. The AI fills the parameters; you choose what it sees.

Explore Operate

Test & Debug

Sandbox conversations, a full debugger for every reply, and an AI diagnose-and-fix loop.

Explore Operate

Rate limits & queue

Provider-aware pacing, smart backoff, horizontal workers and a Queue page that explains failures.

Explore
Get started

Put it to work on your own inbox

Connect an inbox, describe the job, test it in a sandbox and go live with exactly the autonomy you're comfortable with.

Sign up free See pricing