◂ SCOPE CREEPER
LAUNCH · CLAUDE CODE MCP

GIVE YOUR AI AGENT THE ABILITY
TO ASK ANOTHER AI TO KILL ITS PLAN

@scopecreeper/mcpis now on npm. It's a Model Context Protocol server that exposes three Scope Creeper tools — scan, kill, shippable — to Claude Code (or any MCP-speaking client).

The point: Claude is sycophantic by default. Ask it for a plan and you get a plan. Scope Creeper is the opposite — its kill tool returns a brutally honest one-page autopsy arguing why a project should NOT be built, with dated cutoff signals and a one-line eulogy. Install the MCP and your agent gains an adversarial second opinion it can invoke before writing a single line of code.

Install in 30 seconds

(Optional) generate an API key at scopecreeper.ai/account. Without a key the free tier still works — keys just identify your Pro entitlement.

claude mcp add scope-creeper -- npx -y @scopecreeper/mcp \
  --api-key=sk_sc_live_YOUR_KEY

Three tools

scope_creeper_scan(payload) — quick delusion score (0-100) + tier + verdict + 3-5 creep dimensions. Free tier.

scope_creeper_kill(plan)— markdown autopsy with sunk-cost framing, dated cutoff signals, "what to build instead", and a one-line eulogy. Free tier. This is the one that matters.

scope_creeper_shippable(plan) — 1-page PRD with a named stack, V0 scope (3-5 bullets), 3-4 acceptance criteria, and 4-6 paste-runnable shell commands for the first 30 minutes of work. Pro tier ($9/mo).

The workflow

Before any big build, ask Claude Code:

> Use scope-creeper to KILL this plan before we start:
>   "Build a Notion competitor with AI, calendar, CRM, voice notes."

Claude calls scope_creeper_kill and pastes back the autopsy. If the autopsy is convincing, you walked away cheap. If the plan survives the autopsy, ask for scope_creeper_shippable and get a paste-runnable v0 spec.

Why we built it

Scope Creeper started as a web tool. The web is good for discovery and the "score my repo" moment, but the actual decision to start building usually happens inside a chat — exactly where Claude Code lives. The MCP puts the adversarial second opinion at the moment of decision, not after.

The whole package is 4.5KB on the wire, zero external dependencies beyond the official MCP SDK, no telemetry, and the API key (if you choose to use one) only authenticates — we never see your plan text outside the LLM call that produces the artifact.

Try it free, no key

The free tier — scan + kill — works without authentication, subject to a per-IP rate limit. The Pro tier ($9/mo) gives unlimited everything plus shippable and the rest of the web app (share links, deep-audit, projects).

Sources