Question 1

Is Atender MCP-native?

Accepted Answer

Yes. Atender runs an MCP server at v3.atender.com/api/v1/mcp — stateless streamable HTTP, driving the same API the product runs on. Connect any MCP client and 485 tools load automatically. The list you get is cut to what your credential is allowed to do, so a client only ever sees the tools it can actually call.

Question 2

Which MCP clients work with Atender?

Accepted Answer

Any client that speaks the MCP spec: Claude, Claude Code, Codex, Cursor, and custom agents built on the various agent SDKs. There is no Atender-specific plugin to install — point the client at the server URL, authenticate, and go.

Question 3

Do I have to load all the tools at once?

Accepted Answer

No. The tools are grouped into toolsets, one per area — conversations, kb, agent-stacks, cases, incidents, voice, analytics, webhooks and the rest — and a connection can ask for just the ones it needs. Name a toolset that does not exist and the server tells you which ones do.

Question 4

Can I use the platform through the API?

Accepted Answer

The REST API is the same surface the product uses, not a limited export layer — a create or update goes through the same code the app itself calls. Conversations, contacts, agent stacks, knowledge base, handbook, cases, issues, incidents, analytics, channels, voice, SMS, webhooks, members and roles are all callable, each behind its own scope.

Question 5

Can I configure AI agent stacks via the API or MCP?

Accepted Answer

Yes, for the shape of the stack: the stack itself, its specialists, their playbooks and routing rules, the knowledge and handbook each one may read, and the tools they can call. Enable and disable a stack, reorder its prerequisites, and read back the log of changes. The model behind an agent is not something the API sets, and guardrails are readable — you can list the events where one fired and what it did — but not writable.

Question 6

How do coding agents like Claude Code or Codex use Atender?

Accepted Answer

They connect to the MCP server, discover the tool catalogue, and work with your workspace like a teammate: search the knowledge base, draft and publish articles, inspect conversations, configure agent stacks, and wire webhooks. Scoped keys and OAuth consent keep them inside exactly the permissions you grant.

Question 7

How does authentication work for the MCP server?

Accepted Answer

Two ways. Headless setups use an API key with the scopes you choose. People connecting their own client use OAuth 2.1 with PKCE: the browser opens, a workspace owner signs in and approves, and the client receives tokens that can be revoked on the spot — only an owner can authorise a connection, and the approval is narrowed to what that owner actually holds. Either way, every tool call is bounded by the granted scopes.

Question 8

What are scopes, exactly?

Accepted Answer

Named permissions on a key or a connection — contacts:read, conversations:write, kb, webhooks, voice, analytics and around a hundred others, most split into read and write. Three of them have to be ticked on purpose and are never included in full access: importing conversations, reading agent traces, and writing synthetic conversations.

Question 9

How do webhooks work, and how are they secured?

Accepted Answer

You create a subscription over the API and get a signing secret back once; you can rotate it later, and the old one stops working immediately. Twenty-six events fire today, covering conversations, messages, SLA warnings and breaches, cases, knowledge-base articles, handbook procedures and SMS delivery. Every delivery carries X-Webhook-Signature, an HMAC-SHA256 of the exact bytes sent, plus the event and delivery id. Each delivery is attempted once and recorded, and you can read the attempts back per webhook.

Question 10

Are there official SDKs?

Accepted Answer

Not yet. What there is instead is the OpenAPI spec, served at /api/openapi.json and generated from the API itself, so you can produce a client in whatever language you work in. Browsable docs sit at /api/docs and /api/swagger. And since the MCP server already exposes the platform as tools, agents often need no client at all.

Question 11

Can I test without touching production?

Accepted Answer

Yes. A key is either live or sandbox — sa_live_ or sa_test_ — and anything a sandbox key writes is stamped as such, so a dev script pointed at the wrong key is visible rather than silent.

Question 12

What are the API rate limits?

Accepted Answer

Per key, per minute: 60 on starter, 300 on professional, 1000 on enterprise, with matching daily ceilings and a burst allowance on each. A workspace can be moved to a higher tier, or given its own numbers. Every response carries X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset and X-RateLimit-RetryAfter, so a client can pace itself.

Built for the agent era — MCP & API native

Watch a coding agent drive the platform

What's in MCP & API

Every feature, callable

Point any agent at Atender

Agent stacks, configured from code

Events pushed, never polled

AI Agents

Knowledge Base

Automation

Integrations

The platform your agents can drive.

See Atender in action