# Plugins

Meltano takes a modular approach to data engineering in general and EL(T) in particular, where your project and pipelines are composed of plugins of different types, most notably extractors (Singer taps), loaders (Singer targets), transformers (dbt and dbt models), and orchestrators (currently Airflow, with Dagster in development).

Meltano provides the glue to make these components work together smoothly and enables consistent configuration and deployment.

To learn how to manage your project's plugins, refer to the Plugin Management guide.

# Project plugins

In order to use a given package as a plugin in a project, assuming it meets the requirements of the plugin type in question, Meltano needs to know:

  1. where to find the package, typically a pip package identified by its name on PyPI, public or private Git repository URL, or local directory path,
  2. what settings and other capabilities it supports, and finally
  3. what its configuration should be when invoked.

Together, a package's location (1) and the metadata (2) describing it in terms Meltano can understand make up the base plugin description. In your project, plugins extend this description with a specific configuration (3) and a unique name.

This means that different configurations of the same package (base plugin) would be represented in your project as separate plugins with their own unique names, that can be thought of as differently initialized instances of the same class. For example: extractors tap-postgres--billing and tap-postgres--events derived from base extractor tap-postgres, or tap-google-analytics--client-foo and tap-google-analytics--client-bar derived from base extractor tap-google-analytics.

Each plugin in a project can either:

  • inherit its base plugin description from a discoverable plugin that's supported out of the box,
  • define its base plugin description explicitly, making it a custom plugin, or
  • inherit both base plugin description and configuration from another plugin in the project.

To learn how to add a plugin to your project, refer to the Plugin Management guide.

# Discoverable plugins

Base plugin descriptions for many popular extractors (Singer taps), loaders (Singer targets), and other plugins have already been collected by users and contributed to Meltano's index of discoverable plugins, making them supported out of the box.

Discoverable plugins are defined in the discovery.yml manifest, which can be found in the Meltano repository, ships inside the meltano package, and is available at https://www.meltano.com/discovery.yml. If you'd like to use a different (custom) manifest in your project, put a discovery.yml file at the root of your project, or change the discovery_url setting.

To find discoverable plugins, run meltano discover or refer to the lists of Sources and Destinations.

To learn how to add a discoverable plugin to your project using a shadowing plugin definition or inheriting plugin definition, refer to the Plugin Management guide.

# Variants

In the case of various popular data sources and destinations, multiple alternative implementations of Singer taps (extractors) and targets (loaders) exist, some of which are forks of an original (canonical) version that evolved in their own direction, while others were developed independently from the start.

These different implementations and their repositories typically use the same name (tap-<source> or target-<destination>) and may on the surface appear interchangeable, but often vary significantly in terms of exact behavior, quality, and supported settings.

In its index of discoverable plugins, Meltano considers these different implementations different variants of the same plugin, that share a plugin name and other source/destination-specific details (like a logo and description), but have their own implementation-specific variant name and metadata (like capabilities and settings).

Every discoverable plugin has a default variant that is known to work well and recommended for new users, which will be added to your project unless you explicitly select a different one. Users who already have experience with a different variant (or have specific reasons to prefer it) can explicitly choose to add it to their project instead of the default, so that they get the same behavior and can use the same settings as before. If the variant in question is not discoverable yet, it can be added as a custom plugin.

When multiple variants of a discoverable plugin are available, meltano discover will list their names alongside the plugin name.

To learn how to add a non-default variant of a discoverable plugin to your project, refer to the Plugin Management guide.

# Custom plugins

If you'd like to use a package in your project whose base plugin description isn't discoverable yet, you'll need to collect and provide this metadata yourself.

To learn how to add a custom plugin to your project using a custom plugin definition, refer to the Plugin Management guide.

TIP

Once you've got the plugin working in your project, please consider contributing its description to the discovery.yml manifest to make it discoverable and supported out of the box for new users!

# Plugin inheritance

If you'd like to use the same package (base plugin) in your project multiple times with different configurations, you can add a new plugin that inherits from an existing one.

The new plugin will inherit its parent's base plugin description and configuration as if they were defaults, which can then be overridden as appropriate.

To learn how to add an inheriting plugin to your project using an inheriting plugin definition, refer to the Plugin Management guide.

# Types

Meltano supports the following types of plugins:

  • Extractors pull data out of arbitrary data sources.
  • Loaders load extracted data into arbitrary data destinations.
  • Transforms transform data that has been loaded into a database (data warehouse).
  • Models describe the schema of the data being analyzed and the ways different tables can be joined.
  • Dashboards bundle curated Meltano UI dashboards and reports.
  • Orchestrators orchestrate a project's scheduled pipelines.
  • Transformers run transforms.
  • File bundles bundle files you may want in your project.
  • Utilities perform arbitrary tasks provided by pip packages with executables.

# Extractors

Extractors are pip packages used by meltano elt as part of data integration. They are responsible for pulling data out of arbitrary data sources: databases, SaaS APIs, or file formats.

Meltano supports Singer taps: executables that implement the Singer specification.

To learn which extractors are discoverable and supported out of the box, refer to the Sources page or run meltano discover extractors.

# Extras

Extractors support the following extras:

# catalog extra

An extractor's catalog extra holds a path to a catalog file (relative to the project directory) to be provided to the extractor when it is run in sync mode using meltano elt or meltano invoke.

If a catalog path is not set, the catalog will be generated on the fly by running the extractor in discovery mode and applying the schema, selection, and metadata rules to the discovered file.

Selection filter rules are always applied to manually provided catalogs as well as discovered ones.

While this extra can be managed using meltano config or environment variables like any other setting, a catalog file is typically provided using meltano elt's --catalog option.

# How to use

Manage this extra directly in your meltano.yml project file:



 

extractors:
- name: tap-gitlab
  catalog: extract/tap-gitlab.catalog.json

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _catalog <path>

export <EXTRACTOR>__CATALOG=<path>

meltano elt <extractor> <loader> --catalog <path>

# For example:
meltano config tap-gitlab set _catalog extract/tap-gitlab.catalog.json

export TAP_GITLAB__CATALOG=extract/tap-gitlab.catalog.json

meltano elt tap-gitlab target-jsonl --catalog extract/tap-gitlab.catalog.json

# load_schema extra

  • Setting: _load_schema
  • Environment variable: <EXTRACTOR>__LOAD_SCHEMA, e.g. TAP_GITLAB__LOAD_SCHEMA
  • Default: $MELTANO_EXTRACTOR_NAMESPACE, which will expand to the extractor's namespace, e.g. tap_gitlab for tap-gitlab

An extractor's load_schema extra holds the name of the database schema extracted data should be loaded into, when this extractor is used in a pipeline with a loader for a database that supports schemas, like PostgreSQL or Snowflake.

The value of this extra can be referenced from a loader's configuration using the MELTANO_EXTRACT__LOAD_SCHEMA pipeline environment variable. It is used as the default value for the target-postgres and target-snowflake schema settings.

# How to use

Manage this extra directly in your meltano.yml project file:



 

extractors:
- name: tap-gitlab
  load_schema: gitlab_data

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _load_schema <schema>

export <EXTRACTOR>__LOAD_SCHEMA=<schema>

# For example:
meltano config tap-gitlab set _load_schema gitlab_data

export TAP_GITLAB__LOAD_SCHEMA=gitlab_data

# metadata extra

  • Setting: _metadata, alias: metadata
  • Environment variable: <EXTRACTOR>__METADATA, e.g. TAP_GITLAB__METADATA
  • Default: {} (an empty object)

An extractor's metadata extra holds an object describing Singer stream and property metadata rules that are applied to the extractor's discovered catalog file when the extractor is run using meltano elt or meltano invoke. These rules are not applied when a catalog is provided manually.

Stream (entity) metadata <key>: <value> pairs (e.g. {"replication-method": "INCREMENTAL"}) are nested under top-level entity identifiers that correspond to Singer stream tap_stream_id values. These nested properties can also be thought of and interacted with as settings named _metadata.<entity>.<key>.

Property (attribute) metadata <key>: <value> pairs (e.g. {"is-replication-key": true}) are nested under top-level entity identifiers and second-level attribute identifiers that correspond to Singer stream property names. These nested properties can also be thought of and interacted with as settings named _metadata.<entity>.<attribute>.<key>.

Unix shell-style wildcards can be used in entity and attribute identifiers to match multiple entities and/or attributes at once.

Entity and attribute names can be discovered using meltano select --list --all <plugin>.

# How to use

Manage this extra directly in your meltano.yml project file:



 
 
 
 
 
 

extractors:
- name: tap-postgres
  metadata:
    some_stream_id:
      replication-method: INCREMENTAL
      replication-key: created_at
      created_at:
        is-replication-key: true

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _metadata <entity> <key> <value>
meltano config <extractor> set _metadata <entity> <attribute> <key> <value>

export <EXTRACTOR>__METADATA='{"<entity>": {"<key>": "<value>", "<attribute>": {"<key>": "<value>"}}}'

# Once metadata has been set in `meltano.yml`, environment variables can be used
# to override specific nested properties:
export <EXTRACTOR>__METADATA_<ENTITY>_<ATTRIBUTE>_<KEY>=<value>

# For example:
meltano config tap-postgres set _metadata some_stream_id replication-method INCREMENTAL
meltano config tap-postgres set _metadata some_stream_id replication-key created_at
meltano config tap-postgres set _metadata some_stream_id created_at is-replication-key true

export TAP_POSTGRES__METADATA_SOME_STREAM_ID_REPLICATION_METHOD=FULL_TABLE

# schema extra

  • Setting: _schema
  • Environment variable: <EXTRACTOR>__SCHEMA, e.g. TAP_GITLAB__SCHEMA
  • Default: {} (an empty object)

An extractor's schema extra holds an object describing Singer stream schema override rules that are applied to the extractor's discovered catalog file when the extractor is run using meltano elt or meltano invoke. These rules are not applied when a catalog is provided manually.

JSON Schema descriptions for specific properties (attributes) (e.g. {"type": ["string", "null"], "format": "date-time"}) are nested under top-level entity identifiers that correspond to Singer stream tap_stream_id values, and second-level attribute identifiers that correspond to Singer stream property names. These nested properties can also be thought of and interacted with as settings named _schema.<entity>.<attribute> and _schema.<entity>.<attribute>.<key>.

Unix shell-style wildcards can be used in entity and attribute identifiers to match multiple entities and/or attributes at once.

Entity and attribute names can be discovered using meltano select --list --all <plugin>.

If a schema is specified for a property that does not yet exist in the discovered stream's schema, the property (and its schema) will be added to the catalog. This allows you to define a full schema for taps such as tap-dynamodb that do not themselves have the ability to discover the schema of their streams.

# How to use

Manage this extra directly in your meltano.yml project file:



 
 
 
 
 

extractors:
- name: tap-postgres
  schema:
    some_stream_id:
      created_at:
        type: ["string", "null"]
        format: date-time

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _schema <entity> <attribute> <schema description>
meltano config <extractor> set _schema <entity> <attribute> <key> <value>

export <EXTRACTOR>__SCHEMA='{"<entity>": {"<attribute>": {"<key>": "<value>"}}}'

# Once schema descriptions have been set in `meltano.yml`, environment variables can be used
# to override specific nested properties:
export <EXTRACTOR>__SCHEMA_<ENTITY>_<ATTRIBUTE>_<KEY>=<value>

# For example:
meltano config tap-postgres set _metadata some_stream_id created_at type '["string", "null"]'
meltano config tap-postgres set _metadata some_stream_id created_at format date-time

export TAP_POSTGRES__SCHEMA_SOME_STREAM_ID_CREATED_AT_FORMAT=date

# select extra

  • Setting: _select
  • Environment variable: <EXTRACTOR>__SELECT, e.g. TAP_GITLAB__SELECT
  • Default: ["*.*"]

An extractor's select extra holds an array of entity selection rules that are applied to the extractor's discovered catalog file when the extractor is run using meltano elt or meltano invoke. These rules are not applied when a catalog is provided manually.

A selection rule is comprised of an entity identifier that corresponds to a Singer stream's tap_stream_id value, and an attribute identifier that that corresponds to a Singer stream property name, separated by a period (.). Rules indicating that an entity or attribute should be excluded are prefixed with an exclamation mark (!). Unix shell-style wildcards can be used in entity and attribute identifiers to match multiple entities and/or attributes at once.

Entity and attribute names can be discovered using meltano select --list --all <plugin>.

While this extra can be managed using meltano config or environment variables like any other setting, selection rules are typically specified using meltano select.

# How to use

Manage this extra directly in your meltano.yml project file:



 
 
 

extractors:
- name: tap-gitlab
  select:
  - project_members.*
  - commits.*

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _select '["<entity>.<attribute>", ...]'

export <EXTRACTOR>__SELECT='["<entity>.<attribute>", ...]'

meltano select <extractor> <entity> <attribute>

# For example:
meltano config tap-gitlab set _select '["project_members.*", "commits.*"]'

export TAP_GITLAB__SELECT='["project_members.*", "commits.*"]'

meltano select tap-gitlab project_members "*"
meltano select tap-gitlab commits "*"

# select_filter extra

  • Setting: _select_filter
  • Environment variable: <EXTRACTOR>__SELECT_FILTER, e.g. TAP_GITLAB__SELECT_FILTER
  • meltano elt CLI options: --select and --exclude
  • Default: []

An extractor's select_filter extra holds an array of entity selection filter rules that are applied to the extractor's discovered or provided catalog file when the extractor is run using meltano elt or meltano invoke, after schema, selection, and metadata rules are applied.

It can be used to only extract records for specific matching entities, or to extract records for all entities except for those specified, by letting you apply filters on top of configured entity selection rules.

Selection filter rules use entity identifiers that correspond to Singer stream tap_stream_id values. Rules indicating that an entity should be excluded are prefixed with an exclamation mark (!). Unix shell-style wildcards can be used in entity identifiers to match multiple entities at once.

Entity names can be discovered using meltano select --list --all <plugin>.

While this extra can be managed using meltano config or environment variables like any other setting, selection filers are typically specified using meltano elt's --select and --exclude options.

# How to use

Manage this extra directly in your meltano.yml project file:






 
 

extractors:
- name: tap-gitlab
  select:
  - project_members.*
  - commits.*
  select_filter:
  - commits

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _select_filter '["<entity>", ...]'
meltano config <extractor> set _select_filter '["!<entity>", ...]'

export <EXTRACTOR>__SELECT_FILTER='["<entity>", ...]'
export <EXTRACTOR>__SELECT_FILTER='["!<entity>", ...]'

meltano elt <extractor> <loader> --select <entity>
meltano elt <extractor> <loader> --exclude <entity>

# For example:
meltano config tap-gitlab set _select_filter '["commits"]'
meltano config tap-gitlab set _select_filter '["!project_members"]'

export TAP_GITLAB__SELECT_FILTER='["commits"]'
export TAP_GITLAB__SELECT_FILTER='["!project_members"]'

meltano elt tap-gitlab target-jsonl --select commits
meltano elt tap-gitlab target-jsonl --exclude project_members

# state extra

An extractor's state extra holds a path to a state file (relative to the project directory) to be provided to the extractor when it is run as part of a pipeline using meltano elt.

If a state path is not set, the state will be looked up automatically based on the ELT run's Job ID.

While this extra can be managed using meltano config or environment variables like any other setting, a state file is typically provided using meltano elt's --state option.

# How to use

Manage this extra directly in your meltano.yml project file:



 

extractors:
- name: tap-gitlab
  state: extract/tap-gitlab.state.json

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <extractor> set _state <path>

export <EXTRACTOR>__STATE=<path>

meltano elt <extractor> <loader> --state <path>

# For example:
meltano config tap-gitlab set _state extract/tap-gitlab.state.json

export TAP_GITLAB__STATE=extract/tap-gitlab.state.json

meltano elt tap-gitlab target-jsonl --state extract/tap-gitlab.state.json

# Loaders

Loaders are pip packages used by meltano elt as part of data integration. They are responsible for loading extracted data into arbitrary data destinations: databases, SaaS APIs, or file formats.

Meltano supports Singer targets: executables that implement the Singer specification.

To learn which loaders are discoverable and supported out of the box, refer to the Destinations page or run meltano discover loaders.

# Extras

Loaders support the following extras:

# dialect extra

  • Setting: _dialect
  • Environment variable: <LOADER>__DIALECT, e.g. TARGET_POSTGRES__DIALECT
  • Default: $MELTANO_LOADER_NAMESPACE, which will expand to the loader's namespace. Note that this default has been overridden on discoverable loaders, e.g. postgres for target-postgres and snowflake for target-snowflake.

A loader's dialect extra holds the name of the dialect of the target database, so that transformers in the same pipeline and Meltano UI's Analysis feature can determine the type of database to connect to.

The value of this extra can be referenced from a transformer's configuration using the MELTANO_LOAD__DIALECT pipeline environment variable. It is used as the default value for dbt's target setting, and should therefore correspond to a target name in transform/profile/profiles.yml.

# How to use

Manage this extra directly in your meltano.yml project file:



 

loaders:
- name: target-example-db
  dialect: example-db

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <loader> set _dialect <dialect>

export <LOADER>__DIALECT=<dialect>

# For example:
meltano config target-example-db set _dialect example-db

export TARGET_EXAMPLE_DB__DIALECT=example-db

# target_schema extra

  • Setting: _target_schema
  • Environment variable: <LOADER>__TARGET_SCHEMA, e.g. TARGET_POSTGRES__TARGET_SCHEMA
  • Default: $MELTANO_LOAD_SCHEMA, which will expand to the value of the loader's schema setting

A loader's target_schema extra holds the name of the database schema the loader has been configured to load data into (assuming the destination supports schemas), so that transformers in the same pipeline and Meltano UI's Analysis feature can determine the database schema to load data from.

The value of this extra is usually not set explicitly, since its should correspond to the value of the loader's own "target schema" setting. If the name of this setting is not schema, its value can be referenced from the extra's value using $MELTANO_LOAD_<TARGET_SCHEMA_SETTING>, e.g. $MELTANO_LOAD_DESTINATION_SCHEMA for setting destination_schema.

The value of this extra can be referenced from a transformer's configuration using the MELTANO_LOAD__TARGET_SCHEMA pipeline environment variable. It is used as the default value for dbt's source_schema setting.

# How to use

Manage this extra directly in your meltano.yml project file:





 

loaders:
- name: target-example-db
  settings:
  - name: destination_schema
  target_schema: $MELTANO_LOAD_DESTINATION_SCHEMA # Value of `destination_schema` setting

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <loader> set _target_schema <schema>

export <LOADER>__TARGET_SCHEMA=<schema>

# For example:
meltano config target-example-db set _target_schema '$MELTANO_LOAD_DESTINATION_SCHEMA'

# If the target schema cannot be determined dynamically using a setting reference:
meltano config target-example-db set _target_schema explicit_target_schema

export TARGET_EXAMPLE_DB__TARGET_SCHEMA=explicit_target_schema

# Transforms

Transforms are dbt packages containing dbt models, that are used by meltano elt as part of data transformation.

Together with the dbt transformer, they are responsible for transforming data that has been loaded into a database (data warehouse) into a different format, usually one more appropriate for analysis.

When a transform is added to your project using meltano add, the dbt package Git repository referenced by its pip_url will be added to your project's transform/packages.yml and the package will be enabled in transform/dbt_project.yml.

# Extras

Transforms support the following extras:

# package_name extra

  • Setting: _package_name
  • Environment variable: <TRANSFORM>__PACKAGE_NAME, e.g. TAP_GITLAB__PACKAGE_NAME
  • Default: $MELTANO_TRANSFORM_NAMESPACE, which will expand to the transform's namespace, e.g. tap_gitlab for tap-gitlab

A transform's package_name extra holds the name of the dbt package's internal dbt project: the value of name in dbt_project.yml.

When a transform is added to your project using meltano add, this name will be added to the models dictionary in transform/dbt_project.yml.

The value of this extra can be referenced from a transformer's configuration using the MELTANO_TRANSFORM__PACKAGE_NAME pipeline environment variable. It is included in the default value for dbt's models setting: $MELTANO_TRANSFORM__PACKAGE_NAME $MELTANO_EXTRACTOR_NAMESPACE my_meltano_model.

# How to use

Manage this extra directly in your meltano.yml project file:




 

transforms:
- name: dbt-facebook-ads
  namespace: tap_facebook
  package_name: facebook_ads

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <transform> set _package_name <name>

export <TRANSFORM>__PACKAGE_NAME=<name>

# For example:
meltano config dbt-facebook-ads set _package_name facebook_ads

export DBT_FACEBOOK_ADS__PACKGE_NAME=facebook_ads

# vars extra

  • Setting: _vars
  • Environment variable: <TRANSFORM>__VARS, e.g. TAP_GITLAB__VARS
  • Default: {} (an empty object)

A transform's vars extra holds an object representing dbt model variables that can be referenced from a model using the var function.

When a transform is added to your project using meltano add, this object will be used as the dbt model's vars object in transform/dbt_project.yml.

Because these variables are handled by dbt rather than Meltano, environment variables can be referenced using the env_var function instead of $VAR or ${VAR}.

# How to use

Manage this extra directly in your meltano.yml project file:



 
 

transforms:
- name: tap-gitlab
  vars:
    schema: '{{ env_var(''DBT_SOURCE_SCHEMA'') }}'

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <transform> set _vars <key> <value>

export <TRANSFORM>__VARS='{"<key>": "<value>"}'

# For example
meltano config --plugin-type=transform tap-gitlab set _vars schema "{{ env_var('DBT_SOURCE_SCHEMA') }}"

export TAP_GITLAB__VARS='{"schema": "{{ env_var(''DBT_SOURCE_SCHEMA'') }}"}'

# Models

Models are pip packages used by Meltano UI to aid in data analysis. They describe the schema of the data being analyzed and the ways different tables can be joined, and are used to automatically generate SQL queries using a point-and-click interface.

# Dashboards

Dashboards are pip packages bundling curated Meltano UI dashboards and reports.

When a dashboard is added to your project using meltano add, the bundled dashboards and reports will automatically be added to your project's analyze directory as well.

# Orchestrators

Orchestrators are pip packages responsible for orchestrating a project's scheduled pipelines.

Meltano supports Apache Airflow out of the box, but can be used with any tool capable of reading the output of meltano schedule list --format=json and executing each pipeline's meltano elt command on a schedule.

When the airflow orchestrator is added to your project using meltano add, its related file bundle will automatically be added as well.

# Transformers

Transformers are pip packages used by meltano elt as part of data transformation. They are responsible for running transforms.

Meltano supports dbt and its dbt models out of the box.

When the dbt transformer is added to your project using meltano add, its related file bundle will automatically be added as well.

# File bundles

File bundles are pip packages bundling files you may want in your project.

When a file bundle is added to your project using meltano add, the bundled files will automatically be added as well. The file bundle itself will not be added to your meltano.yml project file unless it contains files that are managed by the file bundle and to be updated automatically when meltano upgrade is run.

# update extra

  • Setting: _update
  • Environment variable: <BUNDLE>__UPDATE, e.g. DBT__UPDATE
  • Default: {} (an empty object)

A file bundle's update extra holds an object mapping file paths (of files inside the bundle, relative to the project root) to booleans.

When a file path's value is True, the file is considered to be managed by the file bundle and updated automatically when meltano upgrade is run.

# How to use

Manage this extra directly in your meltano.yml project file:



 
 

files:
- name: dbt
  update:
    transform/dbt_project.yml: false

Alternatively, manage this extra using meltano config or an environment variable:

meltano config <bundle> set _update <path> <true/false>

export <BUNDLE>__UPDATE='{"<path>": <true/false>}'

# For example:
meltano config --plugin-type=files dbt set _update transform/dbt_project.yml false

export DBT__UPDATE='{"transform/dbt_project.yml": false}'

# Utilities

If none of the other plugin types address your needs, any pip package that exposes an executable can be added to your project as a utility:

meltano add --custom utility <plugin>

# For example:
meltano add --custom utility yoyo
(namespace): yoyo
(pip_url): yoyo-migrations
(executable): yoyo

You can then invoke the executable using meltano invoke:

meltano invoke <plugin> [<executable arguments>...]

# For example:
meltano invoke yoyo new ./migrations -m "Add column to foo"

The benefit of doing this as opposed to adding the package to requirements.txt or running pip install <package> directly is that any packages installed this way benefit from Meltano's virtual environment isolation. This avoids dependency conflicts between packages.