Skip to content
This repository has been archived by the owner on Mar 4, 2022. It is now read-only.
/ prototool Public archive

Your Swiss Army Knife for Protocol Buffers


Notifications You must be signed in to change notification settings


Repository files navigation


MIT License GitHub Release Build Status Coverage Status Docker Image Homebrew Package AUR Package

Update: We recommend checking out Buf, which is under active development. There are a ton of docs for getting started, including for migration from Prototool.

Protobuf is one of the best interface description languages out there - it's widely adopted, and after over 15 years of use, it's practically bulletproof. However, working with Protobuf and maintaining consistency across your Protobuf files can be a pain - protoc, while being a tool that has stood the test of time, is non-trivial to use, and the Protobuf community has not developed common standards with regards to stub generation. Prototool aims to solve this by making working with Protobuf much simpler.

Prototool lets you:

  • Handle installation of protoc and the import of all of the Well-Known Types behind the scenes in a platform-independent manner.
  • Standardize building of your Protobuf files with a common configuration.
  • Lint your Protobuf files with common linting rules according to Google' Style Guide, Uber's V1 Style Guide, Uber's V2 Style Guide, or your own set of configured lint rules.
  • Format your Protobuf files in a consistent manner.
  • Create Protobuf files from a template that passes lint, taking care of package naming for you.
  • Generate stubs using any plugin based on a simple configuration file, including handling imports of all the Well-Known Types.
  • Call gRPC endpoints with ease, taking care of the JSON to binary conversion for you.
  • Check for breaking changes on a per-package basis, verifying that your API never breaks.
  • Output errors and lint failures in a common file:line:column:message format, making integration with editors possible, Vim integration is provided out of the box.

Prototool accomplishes this by downloading and calling protoc on the fly for you, handing error messages from protoc and your plugins, and using the generated FileDescriptorSets for internal functionality, as well as wrapping a few great external libraries already in the Protobuf ecosystem. Compiling, linting and formatting commands run in around 3/100ths of second for a single Protobuf file, or under a second for a larger number (500+) of Protobuf files.

Table Of Contents


Prototool can be installed on Mac OS X or Linux through a variety of methods.

See for full instructions.

Quick Start

We'll start with a general overview of the commands. There are more commands, and we will get into] usage below, but this shows the basic functionality.

prototool help
prototool lint idl/uber # search for all .proto files recursively, obeying exclude_paths in prototool.yaml or prototool.json files
prototool lint # same as "prototool lint .", by default the current directory is used in directory mode
prototool create foo.proto # create the file foo.proto from a template that passes lint
prototool files idl/uber # list the files that will be used after applying exclude_paths from corresponding prototool.yaml or prototool.json files
prototool lint --list-linters # list all current lint rules being used
prototool lint --list-all-lint-groups # list all available lint groups, currently "google" and "uber"
prototool compile idl/uber # make sure all .proto files in idl/uber compile, but do not generate stubs
prototool generate idl/uber # generate stubs, see the generation directives in the config file example
prototool grpc idl/uber --address --method foo.ExcitedService/Exclamation --data '{"value":"hello"}' # call the foo.ExcitedService method Exclamation with the given data on
prototool descriptor-set --include-imports idl/uber # generate a FileDescriptorSet for all files under idl/uber, outputting to stdout, a given file, or a temporary file
prototool break check idl/uber --git-branch master # check for breaking changes as compared to the Protobuf definitions in idl/uber on the master branch

Full Example

See the example directory.

The make command make example runs prototool while installing the necessary plugins.


Prototool operates using a config file named either prototool.yaml or prototool.json. Only one of prototool.yaml or prototool.json can exist in a given directory. For non-trivial use, you should have a config file checked in to at least the root of your repository. It is important because the directory of an associated config file is passed to protoc as an include directory with -I, so this is the logical location your Protobuf file imports should start from.

Recommended base config file:

  version: 3.11.0
  group: uber2

See for how Prototool handles working with protoc.

The command prototool config init will generate a config file in the current directory with the currently recommended options set.

When specifying a directory or set of files for Prototool to operate on, Prototool will search for config files for each directory starting at the given path, and going up a directory until hitting root. If no config file is found, Prototool will use default values and operate as if there was a config file in the current directory, including the current directory with -I to protoc.

If multiple prototool.yaml or prototool.json files are found that match the input directory or files, an error will be returned.

See etc/config/example/prototool.yaml all available options.

File Discovery

In most Prototool commands, you will see help along the following lines:

$ prototool help lint
Lint proto files and compile with protoc to check for failures.

  prototool lint [dirOrFile] [flags]

dirOrFile can take two forms:

  • You can specify exactly one directory. If this is done, Prototool goes up until it finds a prototool.yaml or prototool.json file (or uses the current directory if none is found), and then uses this config for all .proto files under the given directory recursively, except for files in the excludes lists in prototool.yaml or prototool.json files.
  • You can specify exactly one file. This has the effect as if you specified the directory of this file (using the logic above), but errors are only printed for that file. This is useful for e.g. Vim integration.
  • You can specify nothing. This has the effect as if you specified the current directory as the directory.

The idea with "directory builds" is that you often need more than just one file to do a protoc call, for example if you have types in other files in the same package that are not referenced by their fully-qualified name, and/or if you need to know what directories to specify with -I to protoc (by default, the directory of the prototool.yaml or prototool.json file is used).

Command Overview

Let's go over some of the basic commands.

prototool config init

Create a prototool.yaml file in the current directory with the currently recommended options set.

Pass the --document flag to generate a prototool.yaml file with all other options documented and commented out.

Pass the --uncomment flag to generate prototool.yaml file with all options documented but uncommented.

See etc/config/example/prototool.yaml for the config file that prototool config init --uncomment generates.

prototool compile

Compile your Protobuf files, but do not generate stubs. This has the effect of calling protoc with -o /dev/null.

Pass the --dry-run flag to see the protoc commands that Prototool runs behind the scenes.

prototool generate

Compile your Protobuf files and generate stubs according to the rules in your prototool.yaml or prototool.json file.

See etc/config/example/prototool.yaml for all available options. There are special options available for Golang plugins, and plugins that output a single file instead of a set of files. Specifically, you can output a single JAR for the built-in protoc java plugin, and you can output a file with the serialized FileDescriptorSet using the built-in protoc descriptor_set plugin, optionally also calling --include_imports and/or --include_source_info.

Pass the --dry-run flag to see the protoc commands that Prototool runs behind the scenes.

See example/proto/prototool.yaml for a full example.

prototool lint

Lint rules can be set using the configuration file. See the configuration at etc/config/example/prototool.yaml for all available options. There are three pre-configured groups of rules, the setting of which is integral to the prototool lint, prototool create, and prototool format commands:

  • uber2: This lint group follows the V2 Uber Style Guide, and makes some modifications to more closely follow the Google Cloud APIs file structure, as well as adding even more rules to enforce more consistent development patterns. This is the lint group we recommend using.
  • uber1: This lint group follows the V1 Uber Style Guide. For backwards compatibility reasons, this is the default lint group, however we recommend using the uber2 lint group.
  • google: This lint group follows the Google Style Guide. This is a small group of rules meant to enforce basic naming. The style guide is copied to etc/style/google/google.proto.

The flag --generate-ignores will help with migrating to a given lint group by generating the configuration to ignore existing lint failures on a per-file basis.

See for full instructions.

prototool format

Format a Protobuf file and print the formatted file to stdout. There are flags to perform different actions:

  • -d Write a diff instead.
  • -f Fix the file according to the Style Guide. This will have different behavior if the uber2 lint group is set.
  • -l Write a lint error in the form file:line:column:message if a file is unformatted.
  • -w Overwrite the existing file instead.
prototool create

Create Protobuf files from a template. With the provided Vim integration, this will automatically create new files that pass lint when a new file is opened.

See for full instructions.

prototool files

Print the list of all files that will be used given the input dirOrFile. Useful for debugging.

prototool break check

Protobuf is a great way to represent your APIs and generate stubs in each language you develop with. As such, Protobuf APIs should be stable so as not to break consumers across repositories. Even in a monorepo context, making sure that your Protobuf APIs do not introduce breaking changes is important so that different deployed versions of your services do not have wire incompatibilities.

Prototool exposes a breaking change detector through the prototool break check command. This will check your current Protobuf definitions against a past version of your Protobuf definitions to see if there are any source or wire incompatible changes. Some notes on this command:

  • The breaking change detection operates on a per-package basis, not per-file - definitions can be moved between files within the same Protobuf package without being considered breaking.
  • The breaking change detector can either check against a given git branch or tag, or it can check against a previous state saved with the prototool break descriptor-set command.
  • The breaking change detector understands the concept of beta vs. stable packages, discussed in the V2 Style Guide. By default, the breaking change detector will not check beta packages for breaking changes, and will not allow stable packages to depend on beta packages, however both of these options are configurable in your prototool.yaml file.

See for full instructions.

prototool descriptor-set

Produce a serialized FileDescriptorSet for all Protobuf definitions. By default, the serialized FileDescriptorSet is printed to stdout. There are a few options:

  • --include-imports, --include-source-info are analagous to protoc's --include_imports, --include_source_info flags.
  • --json outputs the FileDescriptorSet as JSON instead of binary.
  • -o writes the FileDescriptorSet to the given output file path.
  • --tmp writes the FileDescriptorset to a temporary file and prints the file path.

The outputted FileDescriptorSet is a merge of all produced FileDescriptorSets for each Protobuf package compiled.

This command is useful in a few situations.

One such situation is with external gRPC tools such as grpcurl or ghz. Both tools take a path to a serialized FileDescriptorSet for use to figure out the request/response structure of RPCs when the gRPC reflection service is not available. prototool descriptor-set can be used to generate these FileDescriptorSets on the fly.

grpcurl -protoset $(prototool descriptor-set --include-imports --tmp) ...
ghz -protoset $(prototool descriptor-set --include-imports --tmp) ...

You can also just save the file once and not re-compile each time.

prototool descriptor-set --include-imports -o descriptor_set.bin
grpcurl -protoset descriptor_set.bin ...
ghz -protoset descriptor_set.bin ...

Another situation is to use jq to make arbitrary queries on your Protobuf definitions.

For example, if your Protobuf definitions are in path/to/proto, the following will print all message names.

prototool descriptor-set path/to/proto --json | \
  jq '.file[] | select(.messageType != null) | .messageType[] | .name' | \
  sort | uniq
prototool grpc

Call a gRPC endpoint using a JSON input. What this does behind the scenes:

  • Compiles your Protobuf files with protoc, generating a FileDescriptorSet.
  • Uses the FileDescriptorSet to figure out the request and response type for the endpoint, and to convert the JSON input to binary.
  • Calls the gRPC endpoint.
  • Uses the FileDescriptorSet to convert the resulting binary back to JSON, and prints it out for you.

See for full instructions.

Tips and Tricks

Prototool is meant to help enforce a consistent development style for Protobuf, and as such you should follow some basic rules:

  • Have all your imports start from the directory your prototool.yaml or prototool.json file is in. While there is a configuration option protoc.includes to denote extra include directories, this is not recommended.
  • Have all Protobuf files in the same directory use the same package.
  • Do not use long-form go_package values, ie use foopb, not;foopb. This helps prototool generate do the best job.

Vim Integration

This repository is a self-contained plugin for use with the ALE Lint Engine. The Vim integration will currently compile, provide lint errors, do generation of your stubs, and format your files on save. It will also optionally create new files from a template when opened.

See for full instructions.


Prototool is generally available, and conforms to SemVer, so Prototool will not have any breaking changes on a given major version, with some exceptions:

  • Commands under the x top-level command are experimental, and may change or be deleted between minor versions of Prototool. We expect such commands to be promoted to stable within a few minor releases, however development is still in-progress.
  • The output of the formatter may change between minor versions. This has not happened yet, but we may change the format in the future to reflect things such as max line lengths.
  • The breaking change detector's output format currently does not output filename, line, or column. This is an expected upgrade in the future, so the output will likely change. This is viewed as purely an upgrade, so until this is done, do not parse prototool break check output in scripts.
  • The breaking change detector may have additional checks added between minor versions, and therefore a change that might not have been breaking previously might become a breaking change. This may become stable in the near future, and at this time we'll denote that no more checks will be added.


See for concerns related to Prototool development.

See for maintenance-related tasks.


See for answers to frequently asked questions.

Special Thanks

Prototool uses some external libraries that deserve special mention and thanks for their contribution to Prototool's functionality:

  • - The Golang Protobuf parsing library that started it all, and is still used for the linting and formatting functionality. We can't thank Ernest Micklei enough for his help and putting up with all the filed issues.
  • - Used for the JSON to binary and back conversion. Josh Humphries is an amazing developer, thank you so much.
  • - Still used for the gRPC functionality. Again a thank you to Josh Humphries and the team over at FullStory for their work.