Granary is a library and REST API that fetches and converts between a wide variety of data sources and formats:

Free yourself from silo API chaff and expose the sweet social data foodstuff inside in standard formats and protocols!

Here’s how to get started:

License: This project is placed in the public domain.


All dependencies are handled by pip and enumerated in requirements.txt. We recommend that you install with pip in a virtualenv. (App Engine details.)

The library and REST API are both based on the OpenSocial Activity Streams service.

Let’s start with an example. This code using the library:

from granary import twitter

is equivalent to this HTTP GET request:

They return the authenticated user’s Twitter stream, ie tweets from the people they follow. Here’s the JSON output:

  "itemsPerPage": 10,
  "startIndex": 0,
  "totalResults": 12,
  "items": [{
      "verb": "post",
      "id": ",2013:374272979578150912",
      "url": "",
      "content": "Getting stuff for barbecue tomorrow. No ribs left! Got some nice tenderloin though. (@ Metro Plus Famille Lemay)",
      "actor": {
      "username": "evanpro",
        "displayName": "Evan Prodromou",
        "description": "Prospector.",
        "url": "",
      "object": {
        "tags": [{
            "url": "",
            "startIndex": 113,
            "length": 22,
            "objectType": "article"
          }, "..."],
    }, "..."]

The request parameters are the same for both, all optional: USER_ID is a source-specific id or @me for the authenticated user. GROUP_ID may be @all, @friends (currently identical to @all), @self, @search, or @blocks; APP_ID is currently ignored; best practice is to use @app as a placeholder.

Paging is supported via the startIndex and count parameters. They’re self explanatory, and described in detail in the OpenSearch spec and OpenSocial spec.

When using the GROUP_ID @search (for platforms that support it — currently Twitter and Instagram), provide a search string via the q parameter. The API is loosely based on the OpenSearch spec, the OpenSocial Core Container spec, and the OpenSocial Core Gadget spec.

Output data is JSON Activity Streams 1.0 objects wrapped in the OpenSocial envelope, which puts the activities in the top-level items field as a list and adds the itemsPerPage, totalCount, etc. fields.

Most Facebook requests and all Twitter, Google+, Instagram, and Flickr requests will need OAuth access tokens. If you’re using Python on Google App Engine, oauth-dropins is an easy way to add OAuth client flows for these sites. Otherwise, here are the sites’ authentication docs: Facebook, Flickr, Google+, Instagram, Twitter.

If you get an access token and pass it along, it will be used to sign and authorize the underlying requests to the sources providers. See the demos on the REST API endpoints above for examples.

Using the REST API

The endpoints above all serve the OpenSocial Activity Streams REST API. Request paths are of the form:


All query parameters are optional. FORMAT may be json (the default), xml, or atom, both of which return Atom. atom supports a boolean reader query parameter for toggling rendering appropriate to feed readers, e.g. location is rendered in content when reader=true (the default). The rest of the path elements and query params are described above.

Errors are returned with the appropriate HTTP response code, e.g. 403 for Unauthorized, with details in the response body.

By default, responses are cached and reused for 5m without re-fetching the source data. (Instagram responses are cached for 60m.) You can prevent this by adding the cache=false query parameter to your request.

To use the REST API in an existing ActivityStreams client, you’ll need to hard-code exceptions for the domains you want to use e.g., and redirect HTTP requests to the corresponding endpoint above.

The web UI ( currently only fetches Facebook access tokens for users. If you want to use it to access a Facebook page, you’ll need to get an access token manually with the Graph API Explorer (click on the Get To… drop-down) . Then, log into Facebook on and paste the page access token into the access_token text box.

(Google+ pages aren’t supported in their API.)

Using the library

See the example above for a quick start guide.

Clone or download this repo into a directory named granary (note the underscore instead of dash). Each source works the same way. Import the module for the source you want to use, then instantiate its class by passing the HTTP handler object. The handler should have a request attribute for the current HTTP request.

The useful methods are get_activities() and get_actor(), which returns the current authenticated user (if any). See the individual method docstrings for details. All return values are Python dicts of decoded ActivityStreams 1 JSON.

The microformats2.*_to_html() functions are also useful for rendering ActivityStreams 1 objects as nicely formatted HTML.


Check out the oauth-dropins Troubleshooting/FAQ section. It’s pretty comprehensive and applies to this project too. For searchability, here are a handful of error messages that have solutions there:

bash: ./bin/easy_install: ...bad interpreter: No such file or directory

ImportError: cannot import name certs

ImportError: cannot import name tweepy

File ".../site-packages/tweepy/", line 68, in _get_request_token
  raise TweepError(e)
TweepError: must be _socket.socket, not socket

Future work

We’d love to add more sites! Off the top of my head, YouTube, Tumblr,, Sina Weibo, Qzone, and RenRen would be good candidates. If you’re looking to get started, implementing a new site is a good place to start. It’s pretty self contained and the existing sites are good examples to follow, but it’s a decent amount of work, so you’ll be familiar with the whole project by the end.


Pull requests are welcome! Feel free to ping me with any questions.

You’ll need the App Engine Python SDK version 1.9.15 or later (for vendor support). Add it to your $PYTHONPATH, e.g. export PYTHONPATH=$PYTHONPATH:/usr/local/google_appengine, and then run:

virtualenv local
source local/bin/activate
pip install -r requirements.txt
python test

If you send a pull request, please include (or update) a test for the new functionality if possible! The tests require the App Engine SDK or the Google Cloud SDK (aka gcloud) with the gcloud-appengine-python and gcloud-appengine-python-extras components.

If you want to work on oauth-dropins at the same time, install it in “source” mode with pip install -e <path to oauth-dropins repo>.

To deploy:

python -m unittest discover && gcloud -q app deploy granary-demo *.yaml

To deploy facebook-atom, twitter-atom, instagram-atom, and plusstreamfeed after a granary change:

foreach s (facebook-atom twitter-atom instagram-atom plusstreamfeed)
  cd ~/src/$s && gcloud -q app deploy $s *.yaml

The docs are built with Sphinx, including apidoc, autodoc, and napoleon. Configuration is in docs/ To build them, first install Sphinx with pip install sphinx. (You may want to do this outside your virtualenv; if so, you’ll need to reconfigure it to see system packages with virtualenv --system-site-packages local.) Then, run docs/

This ActivityStreams validator is useful for manual testing.


1.12 - 2018-03-24

  • Add Python 3 support! Granary now requires either Python 2.7+ or Python 3.3+.
  • Instagram:
    • Fix scraping profile pages.
  • Twitter:
    • Update character counting to handle Twitter change that now auto-links all ccTLDs. Background.
  • GitHub:
    • Bug fix for get_activities() with deleted issues and repos.
  • microformats2:
    • object_to_json(): convert tags to simple strings in the category property, not full nested objects like h-cards (#141).
    • Special case GitHub issues that are in-reply-to a repo or its /issues URL to be objectType issue.
    • Render simple string categories in HTML output.

This release is intentionally small and limited in scope to contain any impact of the Ptython 3 migration. It should be a noop for existing Python 2 users, and we’ve tested thoroughly, but I’m sure there are still bugs. Please file issues if you notice anything broken!

1.11 - 2018-03-09

  • Add GitHub!
    • get_activities() supports issues and pull requests, including comments and reactions. It’s currently based on notifications, so it’s best effort, not comprehensive, and only includes recently active issues/PRs.
    • create() and preview_create() support issues, comments, stars, and reactions.
  • Twitter:
    • Prefer MP4 and other video/… content types to HLS (.m3u8) etc. Background.
    • Prefer HTTPS URLs for media images.
    • get_activities(): Support @-prefixed usernames in user_id.
  • Facebook:
    • Support new recurring aka multi-instance events. create() and preview_create() now only support RSVPs to individual instances of multi-instance events, to match the Facebook API itself.
    • Try harder to find original (full) sized photo URLs, specifically _o.jpg files instead of _s.jpg.
    • create() bug fix for photo and image URLs with unicode characters.
    • Fixed bug where get_activities(user_id=...) included the authenticated user’s own recent photos, albums, and news publishes.
  • Instagram:
    • Extract more user (author) data from scraped profile pages.
    • Fix home page feed scraping.
  • microformats2, Atom:
    • Add enclosures for image attachments.
    • Bug fixes for rendering image, video, and audio attachments inside shares and attachments. De-dupe images.
  • microformats2:
    • Handle simple string-only author properties.
    • Add fetch_mf2 kwarg to json_to_object() for fetching additional pages over HTTP to determine authorship.
    • Generate explicit blank p-name in HTML to prevent old flawed implied p-name handling (#131).
    • Fix share verb handling in activity_to_json() and activities_to_html() (#134).
    • Remember which content contains HTML, preserve newlines in it, and don’t translate those newlines to <br>s (#130).
  • Atom:
    • Fix timezone bugs in updated and published.
  • JSON Feed:
    • Omit title from items if it’s the same as the content. (Often caused by microformats2’s implied p-name logic.)

1.10 - 2017-12-10

  • Moved web site and REST API to! now 301 redirects.
  • Twitter:
  • Instagram:
    • Add cookie query param to REST API to allow scraping that logged in user’s feed.
  • HTML (including Atom content):
    • Render image, video, and audio attachments more often and consistently.
    • Include microformats2 u-photo, u-video, and u-audio classes more often and consistently.
  • Atom:
    • Add atom_to_activities() for converting full feed documents.
    • Add to REST API and web UI.
  • JSON Feed:
    • Fix bug that omitted title in some cases (#122).

1.9 - 2017-10-24

  • Add ActivityStreams 2.0! New as2 module includes to_as1() and from_as1() functions. Currently supported: articles, notes, replies, likes, reposts, events, RSVPs, tags, attachments.
  • Atom:
    • Add new atom_to_activity() function for converting Atom to AS1.
    • Add email field to author, if provided.
  • JSON Feed:
    • Raise ValueError on bad (non-dict) input.
    • Add as2 value for format and input. Revise existing ActivityStreams and microformats2 value names to as1, as1-xml, and mf2-json. Old values activitystreams, json, json-mf2, and xml are still accepted, but deprecated.

1.8 - 2017-08-29

  • Add JSON Feed support to both library and REST API.

  • Twitter:

    • Add get_blocklist().
    • Bug fix for creating replies, favorites, or retweets of video URLs, e.g. .
    • Bug fix for parsing favorites HTML to handle a small change on Twitter’s side.
    • post_id() now validates ids more strictly before returning them.
  • Facebook:

    • Improve heuristic for determining privacy of wall posts from other users.
    • Support GIFs in comments (attachment types animated_image_autoplay and animated_image_share).
    • Upgrade Graph API from v2.6 to v2.10.
  • Instagram:

    • Update scraping to handle new home page (ie news feed) JSON schema, which changed sometime around 2017-02-27. (Profile pages and individual photo/video permalinks still haven’t changed yet.)
  • microformats2:

    • Add u-featured to ActivityStreams image.

    • Improve h-event support.

    • Minor whitespace change (added

      ) when rendering locations as HTML.

    • post_id() now validates ids more strictly before returning them.

    • Fix bugs in converting latitude and longitude between ActivityStreams and mf2.

  • Google+:

    • Update HTML scraping to handle changed serialized JSON data format.
  • Atom:

    • Add new activity_to_atom() function that renders a single top-level <entry> instead of <feed>.
    • Add new reader query param for toggling rendering decisions that are specific to feed readers. Right now, just affects location: it’s rendered in the content when reader=true (the default), omitted when reader=false.
    • Include author name when rendering attached articles and notes (e.g. quote tweets).
    • Only include AS activity:object-type and activity:verb elements when they have values.
    • Render AS image and mf2 u-photo if they’re not already in content.
    • Render thr:in-reply-to from object.inReplyTo as well as activity.context.inReplyTo.

    • Fix bugs in html => json-mf2 and html => html conversions.
  • Upgrade brevity to 0.2.14 for a couple bug fixes.

1.7 - 2017-02-27

  • microformats2:
    • Interpret h-cite and u-quotation-of (experimental) as attachments, e.g. for quote tweets.
    • Convert audio and video properties to AS attachments.
  • Twitter:
    • Linkify @-mentions and hashtags in preview_create().
    • Support creating quote tweets from attachments with Twitter URLs.
    • When converting quote tweets to AS, strip quoted tweet URL from end of text.
    • Raise ValueError when get_activities() is passed group_id='@search' but not search_query.
  • Instagram:
  • Facebook:
    • Disable creating “interested” RSVPs, since Facebook’s API doesn’t allow it.
  • Atom:
  • Source.get_activities(): start raising ValueError on bad argument values, notably invalid Facebook and Twitter ids and Instagram search queries.
  • Fix rendering and linkifying content with Unicode high code points (ie above the 16-bit Basic Multilingual Plane), including some emoji, on “narrow” builds of Python 2 with --enable-unicode=ucs2, which is the default on Mac OS X, Windows, and older *nix.

1.6 - 2016-11-26

  • Twitter:
    • Handle new “extended” tweets with hidden reply-to @-mentions and trailing URLs for media, quote tweets, etc. Background:
    • Bug fix: ensure is a plain unicode string so that it can be pickled normally, e.g. by App Engine’s memcache.
    • Bug fix: handle names with emoji correctly in favorites_html_to_likes().
    • Bug fix: handle search queries with unicode characters.
  • Atom:
    • Render full original quoted tweet in retweets of quote tweets.
  • microformats2 HTML:
    • Optionally follow and fetch rel=“author” links.
    • Improve mapping between microformats2 and ActivityStreams ‘photo’ types. (mf2 ‘photo’ type is a note or article with a photo, but AS ‘photo’ type is a photo. So, map mf2 photos to underlying type without photo.)
    • Support location properties beyond h-card, e.g. h-adr, h-geo, u-geo, and even when properties like latitude and longitude appear at the top level.
  • Error handling: return HTTP 502 for non-JSON API responses, 504 for connection failures.

1.5 - 2016-08-25

    • Support tag URI for user id, app id, and activity id.
  • Twitter:
    • Better error message when uploading a photo with an unsupported type.
    • Only include original quote tweets, not retweets of them.
    • Skip fetching retweets for protected accounts since the API call always 403s.
  • Flickr:
    • Better username detection. Flickr’s API is very inconsistent about username vs real name vs path alias. This specifically detects when a user name is probably actually a real name because it has a space.
    • Uploading: detect and handle App Engine’s 10MB HTTP request limit.
    • Bug fix in create: handle unicode characters in photo/video description, hashtags, and comment text.
  • Atom:
    • Bug fix: escape &s in attachments’ text (e.g. quote tweets).

1.4.1 - 2016-06-27

  • Bump oauth-dropins requirement to 1.4.

1.4.0 - 2016-06-27

    • Cache silo requests for 5m by default, 60m for Instagram because they aggressively blocking scraping. You can skip the cache with the new cache=false query param.
  • Facebook:
  • Twitter:
    • Switch create() to use brevity for counting characters.
    • Fix bug in create() that occasionally incorrectly escaped ., +, and - characters.
    • Fix text rendering bug when there are multipl photos/videos.
    • When replying to yourself, don’t add a self @-mention.
  • Instagram:
    • Fix bugs in scraping.
  • Upgrade to requests 2.10.0 and requests-toolbelt 0.60, which support App Engine.

1.3.1 - 2016-04-07

1.3.0 - 2016-04-06

  • Support posting videos! Currently in Facebook, Flickr, and Twitter.
  • Instagram:
  • Facebook:
    • Fetch Open Graph stories aka news.publish actions.
    • Many bug fixes for photo posts: better privacy detection, fix bug that attached comments to wrong posts.
  • Twitter:
    • Handle all photos/videos attached to a tweet, not just the first.
    • Stop fetching replies to @-mentions.
  • Atom:
    • Render attachments.
    • Add xml:base.
  • microformats2:
    • Load and convert h-card.
    • Implement full post type discovery algorithm, using mf2util.
    • Drop support for h-as-* classes, both incoming and outgoing. They’re deprecated in favor of post type discovery.
    • Drop old deprecated u-like and u-repost properties.
  • Misc bug fixes.
  • Set up Coveralls.

1.2.0 - 2016-01-11

  • Improve original post discovery algorithm. (bridgy #51)
  • Flickr tweaks. (bridgy #466)
  • Add mf2, activitystreams, atom, and search to interactive UI. (#31, #29)
  • Improved post type discovery (using mf2util).
  • Extract user web site links from all fields in profile (e.g. description/bio).
  • Add fabricated fragments to comment/like permalinks (e.g. #liked-by-user123) so that object urls are always unique (multiple silos).
  • Improve formatting/whitespace support in create/preview (multiple silos).
  • Google+:
    • Add search.
  • Facebook:
    • Fetch more things in get_activities: photos, events, RSVPs.
    • Support person tags in create/preview.
    • Prevent facebook from automatically consolidating photo posts by uploading photos to “Timeline Photos” album.
    • Include title in create/preview.
    • Improve object id parsing/resolving.
    • Improve tag handling.
    • Bug fix for fetching nested comments.
    • Misc improvements, API error/flakiness handling.
  • Flickr:
    • Create/preview support for photos, comments, favorites, tags, person tags, location.
  • Twitter:
    • Create/preview support for location, multiple photos.
    • Fetch quote tweets.
    • Fetching user mentions improvements, bug fixes.
    • Fix embeds.
    • Misc AS conversion improvements.
  • microformats2:
    • Improve like and repost rendering.
  • Misc bug fixes.
  • Set up CircleCI.

1.1.0 - 2015-09-06

  • Add Flickr.
  • Facebook:
    • Fetch multiple id formats, e.g. with and without USERID_ prefix.
    • Support threaded comments.
    • Switch from /posts API endpoint to /feed.
  • Google+:
    • Support converting HTML to ActivityStreams.
  • Instagram:
    • Support location.
  • Improve original post discovery algorithm.
  • New logo.

1.0.1 - 2015-07-11

  • Bug fix for atom template rendering.
  • Facebook, Instagram: support access_token parameter.

1.0 - 2015-07-10

  • Initial PyPi release.