Skip to main content
The Answering Agent MCP server lets an AI app read your Answering Agent data. Ask “What was our AI-resolved rate by location last week?” or “Find the call where someone mentioned a damaged mirror,” and the app looks up the answer in your account. The app can also change a task’s status or assignee, add internal notes to tasks, and reply to a customer by text or email on an existing conversation. Any app that supports remote MCP (Model Context Protocol) servers can connect, including Claude, ChatGPT, Cursor, and VS Code.
The server runs on Answering Agent, so there is nothing to install. You sign in with your dashboard login over OAuth, and the server sees what you see in the dashboard.
A person asks Claude how many calls the car wash missed last week. Claude runs the Answering Agent run_report tool and answers that the AI finished 195 of 214 calls on its own, with a table of reasons for the 19 that needed a person.A person asks Claude how many calls the car wash missed last week. Claude runs the Answering Agent run_report tool and answers that the AI finished 195 of 214 calls on its own, with a table of reasons for the 19 that needed a person.

Who can connect

  • Team owners, admins, and managers can connect. Other roles see “AI app connections aren’t available for this account.”
  • Each connection works on one team, the team you have open in the dashboard when you connect.

Connect your AI app

These steps work in Claude on the web and in Claude Desktop. A connector you add in one shows up in the other.
1

Add a custom connector

Go to Customize → Connectors. Click + Add, then Add custom connector.The Add menu on Claude's Connectors page is open, with Add custom connector highlighted.The Add menu on Claude's Connectors page is open, with Add custom connector highlighted.
2

Enter the name and URL

Enter Answering Agent as the name and https://answeringagent.com/api/mcp as the URL, then click Continue.Claude's Add custom connector dialog with Answering Agent as the name, https://answeringagent.com/api/mcp as the URL, and the Continue button highlighted.Claude's Add custom connector dialog with Answering Agent as the name, https://answeringagent.com/api/mcp as the URL, and the Continue button highlighted.
3

Keep the detected sign-in settings

Claude selects Sign in now and Use Claude’s published identity, both marked Detected. Leave them as they are and click Add at the bottom of the dialog.The second step of Claude's Add custom connector dialog. Under Authentication, Sign in now is selected and marked Detected. Under OAuth client, Use Claude's published identity is selected and marked Detected. The Add button is highlighted.The second step of Claude's Add custom connector dialog. Under Authentication, Sign in now is selected and marked Detected. Under OAuth client, Use Claude's published identity is selected and marked Detected. The Add button is highlighted.
4

Connect

On the Answering Agent connector page, click Connect.Claude's Answering Agent connector page says You're not connected to Answering Agent yet, with the Connect button highlighted.Claude's Answering Agent connector page says You're not connected to Answering Agent yet, with the Connect button highlighted.
5

Approve the connection

Sign in with your Answering Agent dashboard login if asked. The page names the team you have open in the dashboard. Check the team and the account, then click Approve.The Answering Agent page titled Connect your AI app to Sunny Side Car Wash. It shows the signed-in account, lists what the app will be able to do, and has the Approve button highlighted.The Answering Agent page titled Connect your AI app to Sunny Side Car Wash. It shows the signed-in account, lists what the app will be able to do, and has the Approve button highlighted.
6

Allow access

Team shows the same team. It is the only choice. Click Allow access.The page titled Claude would like access to your account, with Sunny Side Car Wash under Team and the Allow access button highlighted.The page titled Claude would like access to your account, with Sunny Side Car Wash under Team and the Allow access button highlighted.
7

Check tool permissions

Back in Claude, the connector shows Disconnect and its tools. Each tool shows Needs approval, so Claude asks before it runs one.Claude's connected Answering Agent connector with a Disconnect button. Under Tool permissions, the Read-only tools group is set to Needs approval.Claude's connected Answering Agent connector with a Disconnect button. Under Tool permissions, the Read-only tools group is set to Needs approval.
8

Ask a question

Start a new chat and ask about your calls or tasks.
On Team and Enterprise plans, an owner adds the connector first under Organization settings → Connectors. Members then find it under Customize → Connectors and click Connect. Free plans allow one custom connector. See Anthropic’s guide to custom connectors.

Sign in and approve

The first time the app uses the server, it opens a sign-in window.
  1. Sign in with the login you use for the Answering Agent dashboard.
  2. Answering Agent shows Connect your AI app to your team name and lists what the app can do. Check the team, then click Approve.
  3. On the next screen, Team shows the same team. Click Allow access.
  4. Back in the AI app, the server shows as connected. Ask it a question.
If you see This connection request expired, start again from the AI app.

Try these prompts

A person asks ChatGPT for open tasks older than 2 days. ChatGPT calls the Answering Agent search_tasks tool and lists 4 tasks with the customer, age, and assignee.A person asks ChatGPT for open tasks older than 2 days. ChatGPT calls the Answering Agent search_tasks tool and lists 4 tasks with the customer, age, and assignee.
  • “What was our AI-resolved rate each week for the last 8 weeks, by location?”
  • “Who closed the most tasks in the last 30 days, and what was their median time to close?”
  • “What did customers ask for most this week? Group it by category.”
  • “Find the call where a customer mentioned a damaged mirror and summarize what they said.”
  • “How many retention offers did we pitch last month, and how many were accepted?”
  • “Which hours of the day get the most calls at the Lamar Blvd location?”
  • “List the open tasks assigned to me, oldest first.”
  • “What greeting does the Lamar Blvd line use?”
  • “Assign the oldest open damage claim to Priya and add a note that the customer sent photos.”
  • “Mark Dana Ruiz’s cancellation task done and add a note that her membership is cancelled in the POS.”
  • “Reply to the customer on conversation 184532 that their refund is processed.”

Tools

Your AI app picks a tool for each question and sends it to the server. The server reads your team’s data as you and returns the result for the app to answer from. Most tools only look things up: Three tools change data: Only reply_to_conversation contacts a customer. Your AI app decides when to ask before it runs a tool, and Answering Agent does not check whether you approved. By default, Claude and ChatGPT ask you to approve a tool that changes data before it runs. If you set a tool to always run, they stop asking. The MCP server and the API share the lookups for the account, conversations, tasks, contacts, and reports, so a report in Claude matches the same report over HTTP. They also share task updates, task notes, and replies to customers. Only the MCP server has get_settings, list_support_requests, and search_help. Only the API downloads call recordings and lists the metric catalog.

Reply to a customer

reply_to_conversation sends a real text or email. Read the message your AI app shows you before you approve it.
A person asks Claude to reply to the customer on the last cancellation conversation and offer 50% off next month. Claude finds the conversation, drafts the text, and asks for approval to run reply_to_conversation with Deny, Always allow, and Allow once buttons.A person asks Claude to reply to the customer on the last cancellation conversation and offer 50% off next month. Claude finds the conversation, drafts the text, and asks for approval to run reply_to_conversation with Deny, Always allow, and Allow once buttons.
  • The customer receives the reply. It comes from your team’s usual texting number or email address.
  • By default, Claude and ChatGPT ask you before each reply. If you set the tool to always run, replies go out without asking.
  • It replies only on a conversation that already exists. A call can get a text whenever the conversation has a phone number, even before it has a text thread. Otherwise it replies only on a channel that conversation already uses. A website chat with no text or email thread can’t get a reply, and neither can a call with no phone number and no thread. Reply to those from the dashboard.
  • The reply goes only to the conversation’s customer. A text goes to the contact’s current phone number. An email goes to the reply address on the customer’s latest email, or to the contact’s email address.
  • It follows the same rules as the dashboard. You must be allowed to message that customer in the dashboard. A customer who opted out by replying STOP gets no texts. Email still sends, the same as in the dashboard. A text longer than 10 SMS segments is refused.
  • The reply shows in the conversation thread and in Activity under your name, and the conversation counts as handled by a person.
  • If a send fails, the tool says why. The failed message shows in the conversation thread in the dashboard, where you can retry it.
  • Replies have their own limits.

Security

  • You sign in with OAuth. The AI app never sees your password, and you need no API key.
  • The AI app can look up your data, change a task’s status or assignee, add internal notes to tasks, and reply to a customer on an existing conversation. A reply goes only to the conversation’s customer. It cannot change settings.
  • By default, Claude and ChatGPT ask you before they run a tool that changes data, including a reply to a customer. If you set a tool to always run, they stop asking.
  • A connection works on one team and sees what you see in the dashboard. If your access is limited to some phone lines, so is the connection.
  • A change or reply runs as you, with your dashboard permissions. Activity in the dashboard shows it under your name.
  • If you leave the team, or your role changes to one that cannot connect, the connection stops working on its next tool call.
  • To disconnect, remove the server in your AI app.
For hosting, data retention, and compliance status, see Security and data handling.

Connect another team

To use a different team, switch to that team in the dashboard, then connect again. If you switch teams in the dashboard later, the connection for the earlier team stops working until you switch back.

Limits

Each person can make 120 tool calls per minute, and changes count toward the same limit. Replies to customers have their own limits too. One person can send 20 replies per minute, and one conversation can get 5 replies per minute. Past the 120-call limit, the AI app gets “Too many requests. Try again in N seconds.” Past a reply limit, it gets “Too many replies. Try again in N seconds.” The 120-call limit is separate from the API’s.

If something isn’t working

Let your coding agent read these docs

These docs have their own MCP server. It searches and reads the documentation, not your account data, and needs no sign-in. Add it to Claude Code, Cursor, or another coding agent so it can answer questions about the API while you build:
Every page also has a menu at the top to copy it as Markdown or open it in Claude or ChatGPT.