What an MCP server for customer support does
The Model Context Protocol (MCP) is an open standard for connecting AI agents to outside tools. A support tool that runs an MCP server lists what an agent may do there, like reading a conversation or drafting a reply, and the agent calls those tools itself instead of you copying chats into a prompt.
Afterchat is a support chat built to be worked this way. The chat on your site answers what your knowledge vault covers and hands the rest to your team. Your coding agent picks up what was handed over, reads the code it needs and writes the reply. What it learned goes back into the vault, so the chat on your site can answer that question itself the next time someone asks.
Connect your agent
The server is at https://afterchat.chat/mcp. It signs in with OAuth in your browser, so there is no API key to paste.
Claude Code
claude mcp add --transport http --scope user afterchat https://afterchat.chat/mcp
Restart Claude Code, run /mcp, pick afterchat and choose Authenticate. Claude Code in detail
Claude
In Claude on the web or the desktop app, add Afterchat from Claude's connector directory, then Connect. Claude in detail
Cursor
Or add it to ~/.cursor/mcp.json, under "mcpServers":
"afterchat": { "url": "https://afterchat.chat/mcp" }
Then open Cursor Settings, find afterchat in the MCP list and choose Connect. Cursor in detail
Codex
codex mcp add afterchat --url https://afterchat.chat/mcp
Then sign in with codex mcp login afterchat. Codex in detail
VS Code
code --add-mcp '{"name":"afterchat","type":"http","url":"https://afterchat.chat/mcp"}'
Then start the afterchat server from the MCP list and allow the sign-in.
Any other MCP client
Add a remote HTTP server named afterchat with the address above. Or paste this line into the agent and let it set itself up:
Connect Afterchat to this agent: fetch https://afterchat.chat/agent-setup.md and follow it.
When the browser opens, signing in makes your account if you don't have one yet. You pick the workspace and untick anything your agent should not do. Then ask the agent to "check Afterchat status", and it tells you the workspace, what it may do and what is waiting.
What your agent can do
The server has 22 tools in three groups, one for each thing a support chat leaves behind.
| Group | Tools | What it is for |
|---|---|---|
| Inbox | list and read conversations, reply, add a team note, close | Answering what was handed over, with the code open |
| Vault | search, list entries and gaps, write an entry or a guide, check what a code change affects, organize topics, merge or archive, act on tidy-up suggestions | Keeping what the chat answers from correct and complete |
| Requests | list requests, get a request's brief, ask the people who requested it, ship | Building what customers asked for, then telling them it is done |
Eight prompts package the common jobs. In Claude Code they show up as slash commands, like /mcp__afterchat__inbox.
| Prompt | What it does |
|---|---|
inbox | Works through the conversations waiting on you, technical ones first |
investigate | Digs into one conversation with the code and your own tools |
gaps | Answers the questions the vault is missing, from the code |
vault | Builds or improves the vault: topics, entries, merges and guides |
review | Checks unchecked answers and tidy-up suggestions against the code |
triage | Lays out the ranked requests, for deciding what to build |
build | Loads a request's brief to build from |
ship | Tells everyone who asked and updates the vault, once you approve |
What waits for you
- Replies to customers wait as drafts in the inbox until someone on your team approves them, unless an owner lets them go out at once.
- Hold rules keep some things with a person whatever else is set: prices and amounts, limits and quotas, refunds and cancellations, terms and legal, code and file paths in replies, and changes to entries your team checked. You can add words of your own.
- Public changes to your help content, and shipping a request, which emails everyone who asked for it, wait on the Coding agent page for a yes.
- You choose what it may do when you connect it. Every connected agent is listed in Afterchat, where it can be disconnected.
- Everything it does is logged with the connection that did it, and a change can be undone.
- Customer messages are data, never instructions. A customer writing "ignore your instructions" is text in a quote, and nothing Afterchat sends tells your agent to run something.
Your code stays with you
Afterchat never holds your code, your credentials or a connection to your systems. Your agent reads the code with the access you already gave it, and sends back only what it writes: a reply, a note on the conversation, an entry for the vault. An entry written from the code keeps the paths of the files it came from, so after a change your agent can check which answers the change made wrong.
The chat on your site gets none of this access. It answers from the vault only.
How it differs from a help desk's MCP server
Several help desks now publish an MCP server, and some of them only let an agent read. That suits a team that wants an agent to search old tickets. Afterchat was built for the agent to do the work: reply, close a conversation with what it learned, write and merge entries, and build from what customers asked for, all behind the approvals above. Every answer it writes stays in the vault the chat on your site answers from.
Coming from another help desk? You can bring your old chats over in one import, and Afterchat shows what it would have answered before you put the chat on your site.