A commandline tool and Python library for archiving data from Facebook using the Graph API.
Python HTML
Switch branches/tags
Nothing to show
Clone or download
Latest commit 6dc3730 Jan 29, 2018



(The "b" is silent.)

A commandline tool and Python library for archiving data from Facebook using the Graph API.

Facebook data is represented as a graph. The graph is composed of:

  • nodes: Things on Facebook, such as Pages, Albums, and Photos. Each node has an id (e.g., 1322855124437680) and a type (e.g., Page).
  • fields: Attributes such as things, such as name and id.
  • edges: Connections between nodes, e.g., Page's Photos.

The graph is represented as a JSON object. For example:

  "name": "The White House",
  "id": "1191441824276882",
  "about": "Welcome to the official White House Facebook page.

Comments posted on and messages received through White House pages are subject to the Presidential Records Act and may be archived. Learn more at WhiteHouse.gov/privacy.",
  "albums": {
    "data": [
        "created_time": "2017-01-20T19:33:16+0000",
        "name": "Timeline Photos",
        "id": "1199645353456529"
  "metadata": {
    "type": "page"

F(b)arc supports retrieving parts of the graph for archiving. To do so, it allows you to specify what fields and edges to retrieve for a particular node type. (What fields and connections to retrieve is referred to as a definition and is described further below).

Getting API keys

Before you f(b)arc you will need to register an app. To do this:

  1. If you don't already have one, create a Facebook account.
  2. Go to https://developers.facebook.com/apps/ and log in.
  3. Click Add a New App and complete the form.
  4. From the app's dashboard, note the app id and app secret.

See below for more information on tokens.


Note: pip install coming once f(b)arc is more stable.

These are instructions for Python 3. Make appropriate adjustments for Python 2.

  1. Download f(b)arc or clone it:

     git clone https://github.com/justinlittman/fbarc.git
  2. Change to the directory:

     cd fbarc
  3. Optional: Create a virtualenv:

     virtualenv -p python3 ENV
     source ENV/bin/activate
  4. Install requirements:

     pip install -r requirements/requirements3.txt
  5. Get commandline usage:

     python fbarc.py -h



Once you've got your API keys you can tell f(b)arc what they are with the configure command.

python fbarc.py configure

This will store your credentials in a file called .fbarc in your home directory so you don't have to keep providing them. If you would rather supply them directly you can set them in the environment (APP_ID, APP_SECRET) or using commandline options (--app_id, --app_secret).


Using the API requires an access token. F(b)arc supports app access tokens and user access tokens.

F(b)arc can retrieve an app access token using the app id and app secret. However, there are some nodes that cannot be retrieved with an app access token, thus a user access token is recommended.

A user access token allows retrieving more nodes than an app access token (but as used in f(b)arc is still limited to public data). There are two types of user access tokens: short-lived and long-lived tokens. Short-lived access tokens are valid for around an hour; long-lived access tokens for a few months. Long-lived user access tokens are retrieved using a short-lived user access tokens and the app id and app secret.

When given a short-lived access token (e.g., with the configure command), f(b)arc will retrieve and store a long-lived access token. You can get a short-lived access token from https://developers.facebook.com/tools/accesstoken/.

F(b)arc will warn you when you're long-lived user access token is going to expire.


The graph command will retrieve the graph for a node (or use the graphs command to retrieve the graphs for multiple nodes provided in files or stdin). The node is identified by a node id (e.g., 1191441824276882), name (e.g., WhiteHouse) or a Facebook url (e.g., https://www.facebook.com/WhiteHouse/).

The node graph is retrieved according to the specified definition. If the type of a node is not known, provide a definition of discover and f(b)arc will look up the node's type and try to match it to a definition.

f(b)arc finds additional nodes in the graph for a node. For example, for a Page it may find the Album nodes. The --levels parameter will determine the number of levels of nodes that are retrieved, with the default being 1 (i.e., the graph for just the node that was requested). Each additional node graph is returned separately. Setting --levels to 0 will continue until all nodes reachable by edges are exhausted. Be careful, because depending on the definitions, this could be, well, infinite. Use the --exclude parameter to exclude definitions from recursive retrieval.

Note that f(b)arc may need to make multiple requests to retrieve the entire node graph so executing the graph command may take some time.

python fbarc.py graph page 1191441824276882 --levels 2 --pretty

To write the output to a file, use --output-dir or redirect output to a file with > <filename>.jsonl.

python fbarc.py graph page 1191441824276882 --levels 2 --pretty > 1191441824276882.jsonl


The metadata command will retrieve all of the fields and connections for a node.

python fbarc.py metadata 1191441824276882 --pretty

Note that you may not be able to actually retrieve all of those fields or connections with the level of permissions of your API keys. The API will ignore any fields or connections that you cannot access.

The --template and --update parameters help with creating definitions. These are described below.


The url command will return the url for retrieving the graph of a node according to the specified definition.

python fbarc.py url page 1191441824276882


Definitions specify what fields and connections will be returned for a node type, as well as the size of node batches and edges.

Definitions are represented as simple python configuration files stored in the definitions or local_definitions directories. Definitions in definitions are distributed with f(b)arc. You can add additional definitions in local_definitions. A definition in local_definitions with the same filename as a definition in definitions will take precedence.

Here is an example definition for a Page:

definition = {
    'node_batch_size': 10,
    'edge_size': 10,
    'fields': {
        'albums': {'edge_type': 'album'},
        'bio': {},
        likes': {'edge_type': 'page', 'follow_edge': False},
        'name': {'default': True},
        'workflows': {'omit': True},
        'visitor_posts': {'edge_type': 'post', 'omit_on_error': 10}

fields is a map of names to fields or edges to be retrieved for the node.

A name with an edge_type is an edge. The value of edge_type is the name of another definition.

A field or edge in which default is True will always be retrieved. Otherwise, the field or edge will only be retrieved when the node is the primary node being retrieved. In other words, default fields or edges specify the summary for a node type; other fields or edges are part of the detail for a node type.

A field or edge in which omit is True will be ignored. This is helpful for keeping track of fields or edges that have been considered, but are not to be retrieved.

If an edge has follow_edge set to False then only the default fields or edges will be retrieved for that edge. That edge will be omitted from recursive retrieval. For example, for a Page, the likes edge is set to not follow edges because this would cause retrieval of all pages that liked this page, which is not desired.

Sometimes for inexplicable reasons, the Graph API will report errors for particular fields. For example, as of late 2017, requesting the visitor_posts edge on SenatorTedCruz with even a limit of 1 results in a "Please reduce the amount of data you're asking for, then retry your request" error. To handle these sorts of errors, setting omit_on_error will cause the field to be omitted when the specified error is encountered. (Errors are identified using Facebook error codes.)

node_batch_size and edge_size are optional; if omitted sensible defaults will be used. Node batch size determines how many nodes of that type will be requested at a time. A larger number reduces the number of requests to the API, speeding up retrieval. Edge size determines, when retrieving an edge, how many nodes to retrieve. A larger number reduces the number of paging requests, speeding up retrieval. In some cases, limits for node batch size and edge size can be found in the documentation; in others, it must be found by trial and error.

The --template and --update parameters of the metadata command can assist with creating definitions. --template will produce a definition for a node type that includes all possible fields or edges with omit set to True by default. --update will update an existing definition with any new fields or edges that are not already included in the definition. The new field or edges will be indicated by a comment ("Added field") and will have omit set to True.

The Graph API Explorer is helpful for understanding the fields and connections that are available for a node type. Less helpful is the Graph API Reference.

F(b)arc Viewer

F(b)arc Viewer allows you to view and explore the data retrieved from the API.

To run:

python fbarc_viewer.py <filepath(s) of file containing JSON or directories containing JSON files>

Adding --index will cause indexes to be used. Indexes will reduce the amount of memory required. If indexes don't already exist, they will be created:

Once F(b)arc Viewer is running, it will be available at http://localhost:5000/.

Unit tests

To run unit tests:

    python -m unittest discover



Facebook limits retrieving Users. F(b)arc does not support retrieving Users from the graph command, but it does retrieve them when connected from other nodes. The fields that are available are extremely limited.

Incremental archiving

It would be ideal to be able to perform incremental archiving, i.e., only retrieve new or updated nodes. For example, only retrieve new Photos in an Album. Unfortunately, the Graph API doesn't support this. In particular, ordering does not appear to work as documented and if it did work, it is unclear what field is used for ordering.

Suggestions on a strategy for incremental harvesting would be appreciated.

Not yet implemented

  • Search
  • Setup.py
  • Travis configuration


F(b)arc borrows liberally from Twarc in code and spirit.

Facebook policies

Please be mindful of the Facebook Platform Policy.