Choose a public domain and 1–10 posts. The input goes directly to Hacker News.
No provider account or payment setup
02
Use the browser when needed
The engine can fall back from HTTP to a fresh local Chrome profile.
Your personal profile is not imported
03
Receive structured records
Get post IDs, titles and links with a receipt describing execution and validation.
Results stay in your host unless you share them
This helper supports Hacker News only. General web search, account sign-in, article reading, posting and payments are outside this pilot.
REVIEW, CONFIGURE, THEN CONNECT
Set up your local helper.
Unzip the kit into a folder you control. Review its source and instructions before running it. No installation starts from this page.
1. Prepare the engine
Use Node 22.13 or newer. Fetch the pinned upstream revision, install its locked dependencies, and build it. These commands contact GitHub and the package registry.
Upstream MIT license. Chrome must already be installed. This pilot supports macOS and Linux process cleanup. Chrome on macOS was live-tested; Linux was not. Windows is not supported.
2. Configure and check
Run from the extracted kit folder. Replace both placeholder paths with absolute paths on your computer.
doctor checks the local installation. Only call performs the public website request. Your host may ask you to approve local execution.
3. Add the generated MCP configuration
Configuration produces a local MCP configuration file with absolute paths. Merge its server entry into a host that supports mcpServers, or translate the entry into your host’s format. Discover the two tools after connecting.
helper_status → checks the local setup
hn_list_posts → reads public Hacker News posts
Example task: “Use the ToolSkipper local helper to list five Hacker News posts from meta.ai. Return the titles and source links.” Enable the helper only for tasks it supports.
Local MCP integration also passed: a real client retrieved five matching records, received an empty result for a domain with no posts, and passed seven protocol and input-rejection checks. These are internal tests, not external agent adoption.
Plain HTTP returned 419. The engine’s anonymous browser fallback worked. The response fields were configured by a reviewer; this does not demonstrate automatic learning of every website. A successful result is not a guarantee of factual accuracy or future availability.
Keep the useful part local.
Each read uses a temporary browser profile. The helper accepts no provider secrets and does not automatically upload results to ToolSkipper. It includes no paid API calls; your computer, network and agent-model costs may still apply.