Discuria

MCP connector

Think where you already think

Solve real Discuria problems inside Claude, ChatGPT or your coding agent on your own AI subscription, and have the reasoning scored to your portfolio. One sign-in, one permission.

01

How it works

Add the connector
Paste the server URL into your AI app. It registers itself; there is nothing to configure.
Sign in once
Your app opens a Discuria sign-in and asks for one permission: scoring your thinking. Nothing else.
Think, then score
Ask for a problem, work it through in your own words, and say “score my thinking.” The score card lands in your portfolio.

02

MCP endpoint

https://discuria.space/api/mcp

OAuth 2.1 with PKCE · your prompts stay in your AI · only the reasoning you submit for scoring reaches Discuria

03

Set up

Pick your client.

Claude

  1. claude.ai → Settings → Connectors → Add custom connector
  2. Server URL: https://discuria.space/api/mcp
  3. Click Connect. A Discuria sign-in opens; allow scoring.
  4. In any chat: “List Discuria problems.”
  5. Then: “Open the Discuria thinking pad for this one.” A panel opens in the chat (Claude desktop and iOS); type your own steps there and press Score my thinking.

ChatGPT

  1. Settings → Apps & Connectors → Advanced → Developer mode
  2. Create → Server URL: https://discuria.space/api/mcp → Authentication: OAuth
  3. Connect, then allow scoring on the Discuria page that opens.
  4. In a chat, pick the Discuria connector and ask for a problem.
  5. Scoring saves to your portfolio, which needs developer mode with write actions: OpenAI offers that on Business, Enterprise and Edu plans for now. Other plans can still solve here on discuria.space.

Claude Code / Codex / Cursor

  1. Claude Code: claude mcp add --transport http discuria https://discuria.space/api/mcp, then /mcp → Authenticate
  2. Codex (~/.codex/config.toml): [mcp_servers.discuria] url = "https://discuria.space/api/mcp", then codex mcp login discuria
  3. Cursor: add a remote MCP server with the URL; it prompts for sign-in.
  4. Then: “Pick a Discuria problem and walk me through it.”

Local models (LM Studio, Open WebUI, scripts)

  1. Discuria doesn't care which model you think with. A free local model is scored on equal footing with GPT or Claude.
  2. Clients that can't open a sign-in use a personal token: Settings → Connected AI → Create token.
  3. Send it as a header: Authorization: Bearer <token>, with the URL https://discuria.space/api/mcp
  4. Revoke it from the same page whenever you like.

04

What gets scored

Your words, not the model's

The connector asks your AI to send the reasoning steps you actually went through (the questions, the rejected ideas, the checks), not a polished answer it wrote for you. The judge is built to discount rubric vocabulary and hedging that isn't followed by a change of mind, and every score is saved with its trace so it can be reviewed.

Prefer the site? Work a problem in a Discuria workspace, or paste a conversation you already had elsewhere into Capture.

Join the beta Manage connected apps