agentic-facebook 0.6.0

Read your own logged-in Facebook — timelines, feed, comments, search, groups — into clean JSON, over the same GraphQL endpoint your browser uses.

agentic-facebook

Read your own logged-in Facebook — timelines, feed, comments, search, groups — into clean JSON.

---

Facebook has no usable API for reading your own feed. The Graph API was closed to this years ago, and what remains needs app review for permissions it will not grant an individual. The usual workaround — automating a browser and scraping rendered HTML — is slow and breaks with every layout change.

agentic-facebook takes a different route. Facebook's own web client gets its data from a single GraphQL endpoint; this reads that same endpoint using the session your own browser already has. You log in once, by hand, in a real browser. After that the tool stores no password, injects no credentials, and replays nobody else's token — it makes the request your browser would have made.

Read DISCLAIMER.md before using this. Automating a Facebook account violates its Terms of Service, publishing this tool exposes its maintainer, and scraping other people's posts can make you a data controller over their personal data. Use a dedicated/throwaway account, not your primary one.

This is not the first tool that does this. facebook-graphql-scraper captures GraphQL responses via Selenium + selenium-wire with credential-based login. This project's difference is one of degree: it reuses a persisted browser-login profile instead of injecting a username/password, builds on scrapling's actively-maintained fetch stack instead of the largely-unmaintained selenium-wire, and since v0.3.0 reads the GraphQL API over plain HTTP with no browser in the hot path at all.

Features

• Seven composable primitives — fetch (a timeline), about (who someone is), feed, comments, post, search, group. Posts carry url, author_url, and author_id, so one command's output is the next one's input.
• Fast path by default — reads GraphQL over plain HTTP with no browser in the loop, falling back to a real browser automatically when that fails.
• Precise date filtering — --since/--until are a server-side filter, not scroll-until-you-see-it. (Pinned posts and posts whose date could not be located are kept regardless, because neither can be honestly judged against a window.)
• Ranked or chronological, your choice — feed and group default to Facebook's ranking but take --sort recent, because "what's being discussed" and "what was posted since Tuesday" are different questions and a ranked feed can only answer the first.
• Documented output — Post, Comment, Entity and ProfileField, as JSON or NDJSON.
• Self-describing — agentic-facebook catalog prints every command, flag, exit code and object type, generated from the code itself, so it cannot go stale.
• Non-bypassable pacing floors — clamped in code, not asked for in prose. This is what keeps it a personal tool rather than a mass-scraper.

Quick start

Requires Python 3.11+. Install into an isolated environment — this package pins exact Playwright versions through scrapling, so sharing a virtualenv with another Playwright-based tool will break one of them:

uv tool install agentic-facebook     # or: pipx install agentic-facebook
agentic-facebook setup                          # one-time: provisions its own browser
agentic-facebook login                          # opens a real browser — log in by hand
agentic-facebook status                         # exit 0 = ready

Then fetch something. Results are written to a JSON file — only a one-line summary goes to stderr — so pass --output and read the file:

agentic-facebook fetch someone.profile --limit 20 --output posts.json
agentic-facebook about someone.profile --output about.json
agentic-facebook feed --limit 10 --output feed.json
agentic-facebook comments "https://www.facebook.com/someone/posts/pfbid02example" --limit 50 --output comments.json
agentic-facebook search "seoul" --type groups --limit 10 --output groups.json

An explicit --output path is emptied before the run starts, so a failed or zero-result run can never leave an earlier run's data behind — check the exit code before reading the file.

Usage overview

Each command does one thing; chaining them is where the value is. A post's url feeds comments; any author_url feeds fetch or about; a group entity's id feeds group. There is deliberately no crawl command — how deep to go is a judgment for the caller, not a flag.

# who engaged with this post, and what else do they post about?
agentic-facebook comments "<post-url>" --limit 20 --output c.json
#   → read c.json, collect distinct author_url, then:
agentic-facebook fetch "<author_url>" --limit 5 --output person.json
agentic-facebook about "<author_url>" --output whois.json   # …and who they actually are

See Chaining Recipes for worked multi-hop examples, and run agentic-facebook catalog for the authoritative command surface of the version you have installed.

Example output

{
  "id": "ZmVlZGJhY2s6MTIzNDU2Nzg5MDEyMzQ1",
  "url": "https://www.facebook.com/some.profile/posts/pfbid02example",
  "type": "status",
  "author_name": "Jane Example",
  "author_url": "https://www.facebook.com/some.profile",
  "author_id": "100000000000001",
  "created_at": "2026-06-30T09:15:36Z",
  "text": "Full post body, truncation-resolved if it was ever cut short...",
  "media": [],
  "links": [],
  "reaction_count": 370,
  "comment_count": 32,
  "share_count": 14,
  "shared_post": null,
  "source": "timeline",
  "captured_at": "2026-07-05T03:18:13.385206Z"
}

source (timeline | newsfeed | group | search | permalink) means merged results from several commands stay self-describing — permalink being the honest answer for a post fetched directly by URL, whose original surface is unknown. Full field reference: Output Schema, or agentic-facebook schema.

Documentation

Full documentation lives in docs/wiki/ — installation, quick start, architecture, chaining recipes, the complete CLI and Python API references, configuration, troubleshooting, and the security and privacy posture.

Project status

Alpha, pre-1.0, single maintainer. macOS is the tested, first-class platform; Linux likely works for the fetch/parse/CLI layer but is untested against a live session; Windows is unsupported. The output schema is additive-only — new fields are a minor bump; reinterpreting an existing one would be breaking.

Because it depends on Facebook's private GraphQL API, it will break when Facebook ships client changes. fetch falls back to a browser when that happens; the other commands need a package update. See FAQ and Troubleshooting.

Contributing

Issues and pull requests are welcome — see CONTRIBUTING.md. One rule matters more than the rest: never commit captured Facebook data, including as a test fixture.

License

MIT — see LICENSE.