Given a task description and project path, returns the minimum files to read, utilities to reuse, and the callers that are most likely to break.
DESCRIPTION
This is the highest-leverage planning tool in the set. Call it before you start reading files. It tells an agent where to look first and, just as important, which files it can ignore for now.
GOOD FIT
AVOID IT WHEN
| Name | Type | Required | Description |
|---|---|---|---|
| path | string | Yes | Absolute path to the project root. |
| task | string | Yes | Plain-English description of the task you want to do. |
PATH PARAMETER
Use the absolute path to the project root on the machine where the code is available. In Cursor, VS Code, or Windsurf that usually means the workspace root you already have open. On a remote dev box, pass the absolute path on that remote machine. If the path does not exist or the service cannot access it, the tool returns an explicit error instead of guessing. Results are best when the folder is a real project root with .git or package.json. Private GitHub repos are not fetched automatically.
CALL THIS FIRST
Most agents waste 60 to 80 percent of their context reading files that do not matter. scope_task acts as a pre-filter: you describe the work in plain English and it returns only the files you actually need, plus utilities to reuse instead of rebuilding. In practice this usually cuts 3 to 8 unnecessary file reads out of a session.
{ "jsonrpc": "2.0", "id": 8, "method": "tools/call", "params": { "name": "scope_task", "arguments": { "path": "/Users/name/projects/zephex", "task": "Add a rate limit to the /api/webhooks endpoint that allows max 10 requests per minute per IP address" } }}Relevant files:- src/server/webhooks.ts (target route)- src/middleware/rate-limit.ts (existing limiter pattern)- src/lib/redis.ts (shared Redis client)- src/lib/rateLimiter.ts (reusable checkRateLimit helper)Callers at risk: webhook retries, health checksRisk: mediumTOKEN EFFICIENCY
scope_task usually returns about 250 to 600 tokens. Without it, agents often read 4 to 10 files up front, which commonly costs 4,000 to 18,000 tokens before the real work even starts.