Agent Fetch

Fetch web pages as clean Markdown for agent workflows. Uses a fallback pipeline: native Markdown -> static HTML extraction -> headless browser rendering.

When to use agent-fetch

Use agent-fetch instead of (or in addition to) a built-in web-fetch tool when:

• The page is a JavaScript-rendered SPA -- use --mode browser or let auto mode fall back to headless Chrome
• You need custom headers (authentication, cookies) -- use --header (prefer environment variable references such as --header "Authorization: Bearer $TOKEN" to avoid exposing secrets in logs)
• You want the extracted content without AI summarization -- agent-fetch outputs the readability-extracted body directly, subject to --max-body-bytes (default 8 MiB)
• You need to batch fetch multiple URLs concurrently -- pass multiple URLs with --concurrency
• You need to wait for specific content to appear -- use --wait-selector
• No built-in web-fetch tool is available in your environment

Ensure the CLI exists

Require agent-fetch in PATH. If not installed, install it:

Preferred method (no Go required):

1. Download a release artifact from https://github.com/firede/agent-fetch/releases
2. Extract the archive, make the binary executable, and place it on PATH.

macOS note: if Gatekeeper blocks execution, run xattr -dr com.apple.quarantine /path/to/agent-fetch

Alternative method (requires Go):

go install github.com/firede/agent-fetch/cmd/agent-fetch@latest

Verify: agent-fetch --help

Modes

Command patterns

Default (auto mode):

agent-fetch https://example.com

JavaScript-heavy page:

agent-fetch --mode browser --wait-selector "article" https://example.com

Authenticated endpoint:

agent-fetch --header "Authorization: Bearer $TOKEN" https://example.com

Batch fetch:

agent-fetch --concurrency 4 https://example.com https://example.org

Static-only (no browser dependency):

agent-fetch --mode static https://example.com

Without front matter:

agent-fetch --meta=false https://example.com

Diagnose browser readiness:

agent-fetch doctor

Use a custom browser path (useful in containers):

agent-fetch doctor --browser-path /usr/bin/chromium

Structured JSONL output:

agent-fetch --format jsonl https://example.com

Output contract

• Fetched content is written to stdout in the selected format (markdown or jsonl).
• In markdown mode with multiple URLs, output uses task markers:

<!-- count: N, succeeded: X, failed: Y -->
<!-- task[1]: <input-url> -->
...markdown...
<!-- /task[1] -->
<!-- task[2](failed): <input-url> -->
<!-- error[2]: <error text> -->

• In jsonl mode, each task emits one JSON line:

{"seq":1,"url":"https://example.com","resolved_mode":"static","content":"...","meta":{"title":"...","description":"..."}}
{"seq":2,"url":"https://bad.example","error":"http request failed: timeout"}

• Exit code 0: all succeeded. 1: any task failed. 2: argument/usage error.

Error handling