Skip to main content
ZR
Toggle navigation
Work

Flagship mandate

docpull

Local-first context dependency manager for AI agents, MCP clients, and RAG pipelines.

Tier
Flagship mandate
Pillar
OSS infrastructure
Visibility
Public
Status
functional
Date
Oct 2024

Case boundary

Local-first web capture and MCP context infrastructure that turns crawlable pages into source-linked Markdown, JSONL, SQLite FTS5, OKF bundles, and typed context packs.

Claim boundary

Public web and explicitly authorized sources only; browser/cloud paths are explicit escalations.

docpull converts the web into source-linked local context for agent workflows: documentation, reference sites, product pages, changelogs, styleguides, brand profiles, screenshots, schemas, search results, and other crawlable public surfaces.

It is deliberately browser-free by default: async HTTP keeps installs smaller, execution faster, and the crawl surface easier to reason about. Browser rendering, provider-backed search, and paid-capable calls are explicit escalation paths rather than hidden defaults.

pip install docpull
docpull https://example.com

System

  • Web HTML becomes Markdown with YAML frontmatter and source URLs
  • Crawl profiles support RAG context, full mirrors, fast sampling, and typed local packs
  • Pack builders cover brand profiles, styleguides, product pages, schema extraction, images, screenshots, and search
  • Depth, page-count, URL-pattern, cache, and robots controls are explicit
  • Optional JavaScript rendering and provider-backed search paths for sites and workflows that need them

Runtime

  • PyPI CLI for one-off crawls and project setup
  • stdio MCP server for agent-driven web context retrieval
  • Incremental cache profiles with configurable TTLs
  • Markdown, JSONL/NDJSON, SQLite FTS5, OKF bundle, manifest, and context-pack output paths

Proof

  • Blocks private IP ranges, CGNAT, and IPv4-mapped IPv6 SSRF paths
  • Rejects common binary formats before parsing with magic-byte checks
  • Applies URL and file validation before writing crawl artifacts
  • Respects robots.txt and HTTPS-only defaults
  • Makes local, open, browser, provider, and paid-capable boundaries explicit