Read live documentation
Pull Markdown from any docs URL so the agent answers from the current page, not stale training data.
Live web context for AI agents
Give your agent eyes on the web
One MCP install gives your agent live access to the public web. Read docs, capture pricing pages, and extract structured data without writing scrapers.
Works with your agent stack
Set up in under a minute.
One config block. Works with Claude, Cursor, and Codex.
1
2
Pick your preferred AI setup
Use CaptureWeb via MCP, Skills, or Agents.md.
Supports MCP, Skills, and Agents.md.See full agent docs →
3
Paste and go
Run this in your terminal. Your agent reads the docs and handles the rest.
curl -s https://captureweb.dev/agents.md | claude -p "Follow these setup instructions for CaptureWeb"4
You're set
Your agent can fetch Markdown, screenshots, and structured data from any public URL.
Copied!
Use cases
Agents that actually read the internet
Stop pasting URLs into chat. Your agent calls CaptureWeb tools to fetch clean context, then reasons over fresh page content on every turn.
Screenshot + context
Use /snapshot when the agent needs both visual layout and text from the same browser session.
Zero scraper maintenance
JS rendering, bot bypass, and boilerplate removal happen server-side. Your agent code stays thin.
Technical approach
Built for tool-calling agents
CaptureWeb exposes 10 REST endpoints and matching MCP tools. Agents pick the output shape they need without you wiring HTTP in every framework.
MCP-native
Stdio MCP server ships with tool definitions, auth headers, and caching behavior documented in agents.md.
Multi-format tools
markdown, screenshot, pdf, json, brand, and crawl map to distinct tools so agents choose the right abstraction.
Cache-aware
Repeat URLs cost 0 credits. Agents that re-fetch the same docs or pricing page every session stay within free tier limits.
Integration
Connect your agent stack in minutes
Claude Desktop
Add the MCP server to claude_desktop_config.json. Paste agents.md for full endpoint reference.
Setup guide →Cursor & Codex
Open in Cursor or Codex with a pre-built prompt from the code panel. Same cURL as the playground.
Cursor setup →Custom agents
Call the REST API directly from LangChain, n8n, or your own orchestrator. One API key for all endpoints.
Agent overview →001Which MCP tools are available?
Screenshot, PDF, Markdown, HTML content, links, JSON extraction, snapshot, accessibility tree, brand, crawl, and check-links. See agents.md for the full list.
002Do I need the REST API if I use MCP?
No. MCP wraps the same endpoints. Use REST when you want server-side pipelines without an IDE agent attached.
003How fast are tool calls?
Cached URLs return in under 60ms. Uncached renders typically complete in 1 to 3 seconds depending on page complexity.
Explore more
Web capture for every workflow
AI & agents
RAG pipelines
Feed your vector database with fresh, clean web content stripped of boilerplate.
Learn more →AI & agentsAgent navigation
Full ARIA tree as JSON for computer-use agents and automated QA.
Learn more →Data extractionBrand extraction
Turn any company domain into a structured brand profile.
Learn more →Connect your agent today
MCP-ready on day one. 100 free credits to test every tool.