Skip to content

MCP migration: LLM-driven scraping vs fast-path ATS fetchers #12

@Mayank-glitch-cpu

Description

@Mayank-glitch-cpu

Tracking tradeoff from agentic MCP migration plan.

Decision: Ship fast-path tools (fetch_ashby_jobs, fetch_greenhouse_jobs, fetch_lever_jobs, fetch_workday_jobs) disabled by default behind config.enableFastPath.

Open questions:

  • When to recommend users enable fast-path (token cost threshold?)
  • Reliability measurement for web_fetch on JS-heavy ATS (Workday, Ashby)
  • Should Playwright-based Ashby fetcher stay archived forever, or resurrect behind fast-path?
  • Telemetry to compare agent-scraped vs fast-path coverage

Context: Agent + web_search/web_fetch is model-agnostic and maintenance-free but slower and flakier on JS-heavy sites. Existing fetchers are fast but brittle to ATS changes.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions