Skip to content
ISSUE 001·LIVE·13:04 IL

Developers · Agents

Read this journal like a machine.

Everything here is public, read-only and free. No authentication, no signup, no keys. Cache responses for up to 24 hours; new articles arrive on the feed, not by polling pages.

Endpoints

  • /api/articles

    Every article with metadata, English and Hebrew, as JSON.

  • /api/articles/{slug}

    One article with its full Markdown body. Hebrew mirror at /api/he/articles/{slug}.

  • /api/search.json

    Prebuilt search index. Fetch once, match your query against title, description and tags.

  • /openapi.json

    OpenAPI 3.1 description of this API.

  • /api/mcp

    MCP endpoint (streamable HTTP). Tools: list_articles, search_articles, get_article. TEMPORARILY OFFLINE pending hosting cost verification.

  • /llms.txt

    Navigation index for language models. /llms-full.txt carries all article text in one document.

  • /agents/cards/index.json

    Agent site cards: how to use this site, in one JSON file.

  • /feed.xml

    RSS feed. A new article lands about every 12 hours.

Browser tools (WebMCP)

In a browser that implements the WebMCP draft (navigator.modelContext), each tool page below registers its tool for the browser's agent. Execution is the same local code the page uses: inputs never leave the page, nothing is sent to this site, and inputs are capped at 200,000 characters. In any other browser these pages are ordinary interactive tools for people.

  • estimate_task_cost

    On /tools/task-cost. Cost per task and per month from tokens, retries, caching and model routing, with cited list prices.

  • score_credential_check

    On /tools/credential-check. Ten yes/partial/no questions about agent credential storage, scoping, rotation, runtime and audit; returns a score, band and fix list. Never pass a secret value.

  • lint_mcp_config

    On /tools/mcp-lint. Static lint of an MCP config for structural errors and common risk patterns.

  • scan_for_injection

    On /tools/injection-scan. Flags known prompt-injection patterns in text an agent is about to read. Heuristic; a clean result is not a guarantee.

Examples

# List articles
curl https://www.harelasaf.com/api/articles

# One article, full Markdown
curl https://www.harelasaf.com/api/articles/what-is-an-ai-builder

# The same article as Markdown
curl https://www.harelasaf.com/articles/what-is-an-ai-builder.md

# MCP (TEMPORARILY OFFLINE pending hosting cost verification - the endpoint
# currently answers with a static status probe, no JSON-RPC tools):
{
  "mcpServers": {
    "harelasaf-journal": {
      "type": "streamable-http",
      "url": "https://www.harelasaf.com/api/mcp"
    }
  }
}

House rules for agents

  • Read-only. There is no write endpoint, booking flow or form for agents to submit.
  • Quote short passages and cite the canonical article URL you actually read.
  • Prefer /feed.xml or /api/articles to check for new work instead of crawling pages.
  • Contacting the author happens through the channels on /contact, by a person, not in bulk.