Briar

What briar does

The briar MCP server is part of a local command-line application that gathers information from selected engineering and workplace systems and stores the results as markdown knowledge. Supported source categories include GitHub or Bitbucket, Jira or Linear, AWS, and Fireflies. The resulting knowledge can be read, listed, written, or deleted through briar's context commands.

Briar also runs coding workflows. briar agent implement takes a tracker ticket, works in a repository, and can create a branch, run code and tests, and open a draft pull request. briar agent prfix addresses review comments on an existing pull request. Planning commands can turn a Jira or GitHub Projects board into a plan and process its cards sequentially.

How it works

Briar runs on the developer's machine and communicates only with the services configured by the user. There is no hosted briar account or central briar service described in the documentation. Credentials are saved locally, while model access is configured through the selected LLM provider. The base package supports Anthropic and AWS Bedrock; optional extras add OpenAI, Gemini, Google Cloud, Azure, and Vault support.

The MCP functionality is installed with pip install 'briar-cli[mcp]'. That extra provides briar mcp serve, briar chat, MCP tool servers referenced by runbooks, and the think and time helper tools. The server can run over standard input and output, including inside the provided Docker image. A compatible MCP client can start the command as its local tool process.

Knowledge extraction is controlled with commands such as briar extract, which can select extractors and save results to a file or Postgres-backed store. Runbooks can execute extractors once or on a schedule. Configuration values resolve from command-line flags, environment variables, .briar.toml, and built-in defaults, in that order.

Setup and configuration

Install Python 3.10 or newer and the package with the MCP extra:

pip install 'briar-cli[mcp]'

Use briar auth login to save service credentials, then provide the required model-provider credentials through the shell or the configured credential store. briar init --company <name> creates a starter .briar.toml; repository owner and name may also be taken from the Git origin remote. briar config show reports each setting and its source, while briar doctor checks configuration, credentials, Git, and the store.

The Docker image iklob1/briar includes briar, the MCP extra, Git, and the GitHub CLI. Mount the working repository and the local briar configuration directory. For standard-input MCP operation, Docker must be started with interactive input enabled. The image is built for linux/amd64 and runs as user briar with UID 1000.

Tools and capabilities

• Collect knowledge from configured source systems with briar extract.
• Manage stored knowledge blobs with briar context.
• Run ticket implementation and pull-request review-fix workflows.
• Build and execute plans from Jira or GitHub Projects.
• Execute extractor runbooks once or on a schedule.
• Serve MCP tools locally and provide a chat command.

Limitations and notes

The MCP extra is required for briar mcp serve and briar chat; it is also needed when a runbook references MCP tool servers. The README does not document a hosted deployment or a built-in account system. Agent workflows require Git and the GitHub CLI, and real implementations can clone repositories, create branches, commit changes, and open draft pull requests. Docker support for agents configures GitHub authentication, while Bitbucket agents should be run from a local installation.

The briar MCP server does not replace the configured source credentials or LLM provider. Its available information depends on the services, extractors, store, and credentials selected by the user. Docker-mounted directories must be writable by the container user, and Apple Silicon hosts run the published image through emulation.