Connect an agent

Coddle is an MCP server: Claude, Codex and other agents read and edit your schemas and architectures — and for teachers, run lessons too: write drills, check them and give them to classes — straight from the chat. Sign in through OAuth with a consent screen or with a personal token; every edit an agent makes becomes a version carrying its name.

1
Step 1

Server address

One address for every client. Connections and tokens live in your profile.

2
Step 2

Pick where you connect from

Sign in through the browserOAuth

Nothing to copy but the address: Coddle's consent screen opens.

  1. 1Settings → Connectors → “Add custom connector”.
  2. 2Paste https://coddle.uz/mcp and click “Add” — no client secret needed.
  3. 3Coddle's consent screen opens: check the permissions and click “Allow”. Done — Coddle's tools appear in the chat.
What the consent screen shows

The application's name, the return address (localhost for Claude Code and Codex — that is expected) and the requested permissions. Allowing or denying is your call; access can be revoked in your profile at any time.

3
Step 3

Check that it works

Ask the agent: “Show my projects in Coddle”. Every call shows up in the journal in your profile.

Studio: specs, schemas and architectures

What to ask the agent
  • “Write my product up as a spec in Coddle — roles, features with acceptance criteria, data — and build the DB schema from it”
  • “Upload my schema.prisma to Coddle as a new project”
  • “Generate the migration between versions 3 and 5”
  • “Check the orders architecture for 100 000 users and suggest the bottleneck”

Lessons: for teachers

What to ask the agent
  • “Write ten if/else drills for beginners, check them with your own solution, and give them to the class ‘JS, Tuesday’ by Friday.”
  • “Who in my class is stuck on the homework, and on which drill?”

Common questions

OAuth or a token?+

OAuth is easier: nothing to keep, and the access is listed in your profile to revoke. A token is for scripts, CI and machines without a browser.

What does the agent see?+

Only what you see, and only within the rights you gave. Task solutions, battles and ratings are out of its reach.

How do I disconnect an agent?+

Profile → “Connected agents” → “Revoke”. It takes effect at once.

The agent sees only what you see, within the granted permissions. New projects are always private; publishing and deleting need your confirmation. Every call is journaled. Steps checked against the Claude (claude.com/docs/connectors) and Codex (learn.chatgpt.com/docs/extend/mcp) documentation on 22 September 2026.