Contue

Connect your agent

Contue is a remote MCP server. Add it once in your AI tool and sign in with an email code.

Add the server

In your AI tool's connector or MCP settings, add https://getcontue.com/mcp.

Sign in

The tool opens a sign-in page. Enter your email, type the code you get, then choose read and write, or read only.

Use it

Ask your agent to find, read, or update your issues.

For a CLI agent: claude mcp add --transport http contue https://getcontue.com/mcp. Pages for each tool are coming.

For AI agents: how to connect

If a user asked you to connect to Contue, follow this. You do not need anything from the user except one click to approve.

  • Your MCP client can open a browser on the user's device (desktop app, local CLI): just add https://getcontue.com/mcp. It signs in by itself.
  • You run in the cloud, in a sandbox, or as a background task (the sign-in cannot redirect back to you): use the device flow below.
  1. Request a sign-in link:
    curl -X POST https://getcontue.com/oauth/device_authorization -d "client_name=YOUR_AGENT_NAME"
  2. Send the user the verification_uri_complete link from the reply. They open it on any device, sign in with their email code, and tap Allow. The link is valid for 10 minutes.
  3. Meanwhile, every interval seconds (5), ask for the token. authorization_pending means keep waiting; slow_down means wait longer.
    curl -X POST https://getcontue.com/token \
      -d "grant_type=urn:ietf:params:oauth:grant-type:device_code" \
      -d "device_code=DEVICE_CODE" -d "client_id=CLIENT_ID"
  4. When you get access_token, connect to https://getcontue.com/mcp with the header Authorization: Bearer ACCESS_TOKEN. For example:
    claude mcp add --transport http contue https://getcontue.com/mcp --header "Authorization: Bearer ACCESS_TOKEN"

Fallback: the user can also create a personal access token on their account page and give it to you. Tokens last one year and can be revoked there.

How Contue differs from other issue trackers

The same problem, handed off two ways.

Other issue trackers

A team and built-in agents around one issue

Agents live inside the tool. Once assigned, they work alongside people in the comments.

#123New users don’t get the sign-in code
Comments · 8
09:10Member AUser reports no sign-in code
09:32Member AMaybe it’s rate limiting
09:50Member BNot rate limiting: the sending domain isn’t verified
Member B assigns the issue to an agent
10:20AgentVerified the domain, changed the sender
10:35Member CChecked: code arrives in 4 s
11:05Member BJapanese numbers still delayed
11:12Member ACan someone explain the problem in plain words?
11:20Member AJapan can wait until next week
Whoever picks it up, a person or the assigned agent, reads all 8 comments and works out where things stand.
Built for a team moving one thing forward in one tool.

Contue

One person, agents in relay

Work in the agents you already use. One picks up at a time and reads only what hasn’t been folded in.

Q123New users don’t get the sign-in code
Description · current stateUpdated at the 10:41 handoff

Cause: the sending domain wasn’t verified. Domain verified, sender changed; reviewed, code arrives in 4 s.

Timeline
09:10ClaudeUser reports no sign-in codeHanded off
09:32ClaudeMaybe it’s rate limitingHanded off
09:50CodexFound the sending domain unverifiedHanded off
10:20CodexVerified the domain, changed the senderHanded off
10:35ClaudeReviewed: code in 4 sHanded off
Handoff · 10:41
11:05CodexJapanese numbers still delayedNew
11:12You × ClaudeExplain the problem again in plain wordsHistory only
11:20YouJapan can wait until next weekNew
New: still read
History only: skipped
Taking turns:Claude→Codex→Claude→Codex→Anyone nextany agent connected over MCP
The next agent gets the latest description + 2 New entries.
The 5 handed-off entries aren’t re-read, the side chat is skipped, and discarded ideas can’t lead it astray.

The next agent gets only the latest description and what hasn’t been folded in yet.

Say “update progress” whenever you like to keep the description current. Miss one and nothing breaks: New entries still get read.
Agents record everything on their own; you never have to ask.

Recorded automatically via MCP + Hooks