Model Context Protocol

Run your support bots from Claude.

Tendabot ships an MCP server. Connect it once and Claude can read your inbox, rewrite a bot's instructions, fix a wrong answer and tell you what any of it cost — in the same conversation where you noticed the problem.

Start free — 300 credits, no cardSee all 29 tools →

Connecting takes one URL

There is no separate service to run and nothing to install. The server is an endpoint on the platform you already have an account on.

1
Create a key

In the console, go to Settings → Claude & API access and create a key. Choose read-only if you want Claude to look but never touch.

2
Add the server in Claude

Point Claude at the endpoint below. The key travels either in the URL path or as a Bearer header — both work, so use whichever your client supports.

3
Ask for something

Start with “list my bots”. Nearly every other tool takes a botId from that first call, which is exactly what the server tells Claude to do.

Endpoint
https://platform.tendabot.com/api/mcp/YOUR_KEY

# or, with a header instead of the path
https://platform.tendabot.com/api/mcp
Authorization: Bearer YOUR_KEY

What people actually use it for

Manage bots
Six tools cover the whole lifecycle of an agent: find it, read it, change it, and check the change before a customer sees it.
Read more →
Reply to customers
Nine tools for the shared inbox. The important sequence is take_over → send_reply → hand_to_ai: without the first call the AI will answer over the top of you.
Read more →
Add knowledge
Nine tools. The one that earns its keep is delete_knowledge: when a bot gives a wrong answer, the fix is almost always removing a stale chunk, not adding a new one on top.
Read more →
Costs & usage
Two read-only tools, and they answer the question a dashboard cannot: not what did I spend, but which bot on which model is costing more than it is worth.
Read more →
Connect your store
Three tools. The point of sync_catalog is that the bot quotes today’s price and today’s stock, not a snapshot from setup day.
Read more →
Permissions

A read-only key genuinely cannot write.

Every tool declares a scope, and the server checks it before the tool runs — it is not a label on the docs page. Hand Claude a read-only key and a call to send_reply comes back refused, by name, telling you to create a write key if you meant it.

That matters more here than in most integrations, because some of these tools talk to your customers. send_reply posts a real message to a real person.

  • 13 read tools — inbox, bots, knowledge, costs. Safe to hand over.
  • 16 write tools — edits and replies. Refused unless the key allows it.
  • Every call is scoped to the workspaces that key belongs to, and no others
  • Two credential types: an OAuth token from the Connect flow, or a workspace key you create by hand for scripts and CI
  • Deleting a conversation is not exposed at all — some things should need a human, a browser and a confirmation dialog
  • Every call is written to an audit log, including the ones that were refused, so you can see what Claude tried as well as what it did

Questions

Do I need to run a server?

No. It is a route on the platform, not a separate process — nothing to deploy, host or keep alive. If you can reach tendabot.com you can reach the MCP server.

Does it work with anything other than Claude?

It speaks standard MCP over JSON-RPC, so any MCP client can connect. We test against Claude because that is what our customers use, and the setup instructions above are written for it.

What does it cost?

Nothing extra. MCP is included on every plan. The only usage that costs anything is a tool that generates an AI reply — test_bot spends credits because it runs the model. Reading your inbox is free.

Can Claude message my customers by accident?

It can only call send_reply with a write key, and only on a conversation you name. The honest advice: have Claude draft the reply and show it to you first. That is one sentence in your prompt and it removes the risk entirely.

Which workspaces can it see?

Only the ones the key belongs to. A key created in one workspace cannot read another, and an OAuth token resolves your memberships live — so removing someone from a workspace removes their access immediately.

Ready when you are

Your first agent, answering in five minutes.

Create your bot — freeTalk to the demo bot →