Reconstruct MCAP

Takes sequences previously ingested into Mosaico and writes them back out as native .mcap files. It is the inverse operation of the MCAP Ingestion example, which ingests four recordings of the Labelbox robotics datasets from Hugging Face. Connection handling, topic filtering, schema recovery, and file writing are all handled automatically.

Python

Learn

Doc: MCAP Bridge

API Reference: MCAP Bridge

C++

The C++ SDK is currently in development.

Rust

The Rust SDK is currently in development.

Run

mosaicolabs.examples reconstruct_mcaps

Run the command with --help to see all available options.

The daemon must be running, and it must already contain the sequences ingested by the MCAP Ingestion example. This script only reads back data already on the server, it does not download or ingest anything itself.

The full source is on GitHub.

No Adapter Needed to Reconstruct

The MCAP Ingestion example registers no adapter: every channel is stored as an Unmodeled ontology. Reconstruction needs no adapter either. When a sequence is ingested, the original schema and channel of each topic are saved as topic metadata, so the extractor can recover them and rebuild the native messages through the same automatic Unmodeled Adapter used at ingestion time.

If a custom adapter is registered for the topic's schema, its to_mcap() is used instead.

Supported Schema Encodings

The MCAP bridge reconstructs the same schema encodings it ingests:

Schema encoding	Support	Notes
protobuf	Full	Used by every channel of this example.
ros2msg, ros2idl, ros1msg	Not supported	Use the ROS bridge instead, see below.

Support for further encodings will be added in future releases.

note

Reconstruction returns each topic to the schema encoding it was ingested with, as recorded in its metadata: a channel ingested as protobuf is written back as protobuf. Converting a topic to a different schema encoding is not supported.

ROS 2 bags in MCAP format

To reconstruct sequences ingested from ROS 2 bags, use the ROS bridge instead, as shown in the Reconstruct ROS bag example.

Reconstruction Pipeline

MCAPSequenceExtractor is the MCAP specialization of the bridge-agnostic SequenceExtractor, the same base used by the ROS bridge's ROSSequenceExtractor. MCAPExtractorConfig uses the shared ExtractorConfig fields as they are: connection settings, sequence, topic and time-window filters, output path, overwrite, and dry_run.

MCAPSequenceExtractor.run() mirrors the injector's pipeline in reverse:

Connect. Opens a connection to Mosaico using the credentials and TLS settings in MCAPExtractorConfig.

Stream. MosaicoToMCAPLoader reads back every message of the requested sequence, resolving the adapter for each topic as described above.

Write. Translates each message back into its native MCAP encoding and writes it into a fresh .mcap file under saving_path/sequence_name, recreating the original channels and schemas.

Reconstruct every ingested sequence

RECONSTRUCTED_MCAP_FILE_PATH = Path(ASSET_DIR) / "reconstructed"

SEQUENCE_NAMES = [
    "trio_connecting_cables_to_adapters_recording_0001",
    "bimanual_xarm6_place_clothing_into_basket_recording_0001",
    "bimanual_xarm6_remove_hats_and_dirty_laundry_recording_0001",
    "bimanual_xarm6_watering_house_plant_recording_0001",
]

for sequence in SEQUENCE_NAMES:
    configs = MCAPExtractorConfig(
        saving_path=RECONSTRUCTED_MCAP_FILE_PATH,
        sequence_name=sequence,
        host=MOSAICO_HOST,
        port=MOSAICO_PORT,
        log_level=LOG_LEVEL,
        mosaico_api_key=API_KEY,
        enable_tls=ENABLE_TLS,
        overwrite=True,
    )

    extractor = MCAPSequenceExtractor(configs)
    extractor.run()

Each sequence is written to /tmp/mosaico_assets/reconstructed/<sequence_name>/<sequence_name>.mcap, ready to be opened with any standard MCAP tooling (the mcap CLI, Foxglove, ...).