Use Chat from an MCP client
Let an MCP client such as Claude Code start chats in your account, check on them and answer them, while you follow along in Chat.
3 min readUpdated for Chat 0.5.2Edit on GitHub
Connect Claude Code
- 1
Create a token
In Chat, open Settings › Personal, find API access and choose New token. Making a token needs a sign-in from the last 15 minutes; otherwise Chat offers to sign you out and back in. Give it a name you'll recognize, such as the computer it's for, and pick how long it lasts.
- 2
Copy it
Chat shows the token once, with the command that adds it to Claude Code. Copy the command before you close the dialog. Chat keeps only a hash of the token, so it can't show it again.
- 3
Add it to Claude Code
Run the command in a terminal:
claude mcp add --transport http obleth-chat https://chat.example.com/mcp \ --header "Authorization: Bearer obc_…"Then ask Claude Code something like "start three Chat research chats comparing these schedulers, and tell me when they finish".
Other MCP clients connect the same way: Streamable HTTP at https://<your Chat address>/mcp, with the token in an Authorization: Bearer header.
What the client can do
| Tool | What it does |
|---|---|
start_chat | Starts a new chat with a first message, and returns its id and link at once. Optionally picks a model or turns on the chat's workspace first. |
get_chat | Says whether a chat is still running, what it's doing now, any questions it's waiting on, and its latest messages. |
send_message | Sends the next message in a chat that isn't running, such as answers to its questions. |
stop_chat | Stops the reply a chat is writing and keeps what it wrote. |
list_chats | Lists your chats, most recent first, optionally matching a search. |
list_models | Lists the models you can use. |
A chat started this way is an ordinary chat in your history. It runs on the server as if you'd sent the message from the browser, so you can open its link, watch it, and take over at any point. Your daily allowance and the limit on replies running at once apply as usual.
A read-only token gets only get_chat, list_chats and list_models.
What a token can't do
A token works only at /mcp. It can't open the rest of Chat's API, change your settings or account, manage other tokens, approve a tool call that waits for Allow or Deny, or see anyone else's chats. Approvals stay in the browser: a chat that needs one waits until you answer it there.
Revoke a token
Settings › Personal › API access lists your tokens with when each was last used. Revoke stops a token at once. Chats it started keep running.
Chat also ends all of your tokens when you change or reset your password (including with a reset link), when an administrator revokes your sessions or changes your role, and when your account is disabled. Signing out another device, or Sign out other devices, ends the tokens made on that device; tokens made on the device you're using stay.
Limits and security
| Limit | Value |
|---|---|
| Tokens per person | 10 at once |
| Lifetime | 1, 7 or 30 days |
| Calls per token | 120 a minute |
| Text per message returned | 8,000 characters; the chat keeps all of it |
- Making a token needs a sign-in from the last 12 hours. With an older session, Chat asks you to sign in again first.
- Tokens start with
obc_. Chat stores a SHA-256 hash and the first six characters after the marker, which the list shows so you can tell tokens apart. /mcprefuses requests that carry a browserOriginother than Chat's own, so a web page can't use a token it has found.- Creating and revoking tokens, and every chat a token starts, answers or stops, is recorded in the audit log with the token's id.