sparkvibeai · upstream of cursor, claude code & codex

the spec layer your coding agent calls.

Claude Code, Cursor, Codex & Grok commission a calibrated, cross-model-reviewed build spec — scored before you build, without leaving the loop. Connect your agent over MCP, or commission one in the browser.

connect your agent → or commission one here ↓
> work request
⏎ enter to submit · esc to clear

house committee writes the spec · first run free · no card

use-from-your-agent.md MCP

Add one MCP server to Claude Code, Cursor, Codex, or Grok. Your agent commissions a calibrated spec, polls it to a reviewer-derived score, fetches the body, and can push it to a private GitHub repo — without leaving the editor.

mcp config · sparkvibe
{
  "mcpServers": {
    "sparkvibe": {
      "url": "https://api.sparkvibeai.com/mcp",
      "transport": "streamable-http",
      "headers": {
        "Authorization": "Bearer svai.YOUR_CREDENTIAL_ID.YOUR_SECRET"
      }
    }
  }
}
the loop
  1. commission_spec — dispatch the house committee
  2. get_commission — poll to a score
  3. get_spec — pull the spec + quality signals
  4. push_to_github — scaffold a private repo

Read tools too: get_spec_body, get_subscription, and the pre-spec interview (start/answer/skip/get_interview).

Included with Pro and up — generate a key from your account and paste the config. connect from your account → see plans →
recent.specs/

real committee runs, cross-model reviewed and scored before any code — read all three in full on the showcase ▸

how-it-works.md
  1. 1.file a work requesthouse committee drafts candidate specs and scores them
  2. 2.synthesize        one selected spec is produced from the candidates
  3. 3.hand off         the selected spec goes to your coding agent; its reviewer-derived score travels with it
contact.txt