mcp-beaker
MCP server for Beaker lab automation, enabling system provisioning, job management, distro discovery, and failure diagnosis through 25 tools.
README
mcp-beaker
MCP server for Beaker lab automation -- system provisioning, job management, distro discovery, and failure diagnosis.
Works with any Beaker server instance. Built on FastMCP v3 and designed for use with AI coding assistants (Cursor, Claude Desktop, etc.).
Features
- 25 tools covering the full Beaker lifecycle: systems, jobs, distros, tasks
- Flexible auth: Kerberos (native GSSAPI/SPNEGO or
bkrCLI fallback) and password (XML-RPC) - Job XML validation: auto-fills missing fields, infers distro families
- Failure diagnosis: deep analysis with auto-retry on correctable failures
- 10 documentation topics exposed as MCP resources
- 2 workflow prompts for common tasks (reserve system, diagnose job)
- Generic: works with any Beaker URL, configurable SSL/CA settings
Installation
Container (recommended)
The container image bundles everything -- no host dependencies beyond
podman (or docker). Kerberos authentication happens inside the container
via KRB5_PRINCIPAL and KRB5_PASSWORD environment variables.
# Pull the pre-built image from GHCR
podman pull ghcr.io/faizbawa/mcp-beaker:latest
# Or build locally from the repo
podman build -t mcp-beaker:latest -f Containerfile .
Pip / uvx
# Using uv (recommended)
pip install uv
uvx mcp-beaker
# Using pip
pip install mcp-beaker
mcp-beaker
# With native Kerberos support (no bkr CLI needed -- requires krb5-devel on host)
pip install mcp-beaker[kerberos]
# Local development
uv run --directory /path/to/mcp-beaker mcp-beaker
Configuration
Cursor / VS Code
Add to your .cursor/mcp.json (or .vscode/mcp.json):
Container (recommended)
{
"mcpServers": {
"beaker": {
"command": "podman",
"args": [
"run", "--rm", "-i", "--network=host",
"-e", "BEAKER_URL=https://beaker.example.com",
"-e", "KRB5_PRINCIPAL=your-user@YOUR.REALM",
"-e", "KRB5_PASSWORD=your-kerberos-password",
"ghcr.io/faizbawa/mcp-beaker:latest"
]
}
}
}
The container runs kinit internally -- no volume mounts, no host
dependencies. Use --network=host so the container can reach your
Kerberos KDC and Beaker server (especially over VPN).
SSL certificates: Many Beaker servers use internal CA certificates that are not in the container's default trust store. You have three options:
- Disable verification (simplest, fine on a trusted corporate network):
"-e", "BEAKER_SSL_VERIFY=false",
- Pass CA cert inline via
BEAKER_CA_CERT_DATA(no volume mount needed):
"-e", "BEAKER_CA_CERT_DATA=-----BEGIN CERTIFICATE-----\nMIID...your-ca-cert...\n-----END CERTIFICATE-----",
- Mount a CA bundle (single file):
"-v", "/path/to/ca-bundle.crt:/tmp/ca-bundle.crt:ro",
"-e", "BEAKER_CA_CERT=/tmp/ca-bundle.crt",
Pip / uvx
{
"mcpServers": {
"beaker": {
"command": "uvx",
"args": ["mcp-beaker[kerberos]"],
"env": {
"BEAKER_URL": "https://beaker.example.com",
"BEAKER_AUTH_METHOD": "kerberos",
"BEAKER_KERBEROS_BACKEND": "http"
}
}
}
}
Streamable HTTP mode
uvx mcp-beaker --transport streamable-http --port 8000
{
"mcpServers": {
"beaker": {
"url": "http://localhost:8000/mcp",
"type": "streamableHttp"
}
}
}
Environment Variables
| Variable | Required | Default | Description |
|---|---|---|---|
BEAKER_URL |
Yes | -- | Base URL of your Beaker server |
BEAKER_AUTH_METHOD |
No | kerberos |
kerberos or password |
BEAKER_KERBEROS_BACKEND |
No | http |
http (native SPNEGO) or bkr (bkr CLI) |
BEAKER_USERNAME |
For password auth | -- | Beaker username |
BEAKER_PASSWORD |
For password auth | -- | Beaker password |
BEAKER_OWNER |
No | $USER |
Default owner for job queries |
BEAKER_SSL_VERIFY |
No | true |
Verify SSL certificates |
BEAKER_CA_CERT |
No | -- | Path to CA certificate bundle |
KRB5_PRINCIPAL |
Container only | -- | Kerberos principal for kinit inside container |
KRB5_PASSWORD |
Container only | -- | Kerberos password for kinit inside container |
BEAKER_CA_CERT_DATA |
Container only | -- | PEM-encoded CA certificate written to file at startup |
CLI Options
mcp-beaker [OPTIONS]
Options:
--transport [stdio|sse|streamable-http] Transport type (default: stdio)
--port INTEGER Port for HTTP transports (default: 8000)
--host TEXT Host for HTTP transports (default: 0.0.0.0)
--path TEXT Path for streamable-http (default: /mcp)
--beaker-url TEXT Beaker server URL
--ssl-verify / --no-ssl-verify Verify SSL certs (default: verify)
--ca-cert TEXT CA certificate bundle path
--auth-method [kerberos|password] Authentication method
--kerberos-backend [http|bkr] Kerberos backend (default: http)
--read-only Disable all write tools
--enabled-tools TEXT Comma-separated tools to enable
-v, --verbose Increase verbosity (-v info, -vv debug)
--version Show version
--help Show this message
Tools
Read Tools (13)
| Tool | Description |
|---|---|
list_systems |
List systems by availability (all/available/free) |
get_system_details |
Hardware specs, ownership, status for a system |
get_system_history |
Activity history for a system |
get_system_arches |
Supported OS families and architectures |
list_jobs |
Filter jobs by owner, status, whiteboard |
get_job_status |
Job status with failure diagnosis |
get_job_results_xml |
Export job results as XML |
get_job_logs |
List log files for a job/recipe/task |
list_distro_trees |
Search distros by name, family, arch, tags |
list_os_families |
List all known OS families |
whoami |
Show authenticated user info |
list_lab_controllers |
List all lab controllers |
search_tasks |
Search the task library |
Write Tools (12)
| Tool | Description |
|---|---|
submit_job |
Submit a job from XML (with validation and auto-fill) |
clone_job |
Clone and resubmit an existing job |
cancel_job |
Cancel a running/queued job |
watch_job |
Poll until completion with failure analysis and auto-retry |
reserve_system |
Manually reserve a system |
release_system |
Release a manually reserved system |
loan_system |
Grant a loan for a system to a user |
return_loan |
Return a current system loan |
power_system |
Power on/off/reboot a system |
provision_system |
Provision a reserved system with a distro |
extend_watchdog |
Extend a running task's watchdog timer |
set_job_response |
Ack/nak (waive) a recipe set result |
Authentication
Kerberos (recommended)
Ensure you have a valid ticket:
kinit your-username@YOUR.REALM
The server supports two Kerberos backends, controlled by BEAKER_KERBEROS_BACKEND:
| Value | Backend | Install |
|---|---|---|
http (default) |
Native GSSAPI/SPNEGO -- lightweight, pip-installable | pip install mcp-beaker[kerberos] |
bkr |
bkr CLI subprocesses -- traditional, requires RPM |
yum install beaker-client |
Both backends use the same Kerberos ticket from kinit.
Password
Set BEAKER_AUTH_METHOD=password along with BEAKER_USERNAME and BEAKER_PASSWORD. The server authenticates via the XML-RPC auth.login_password() method. Note: this requires server-side LDAP to be enabled.
Architecture
src/mcp_beaker/
__init__.py # Click CLI entry point
config.py # BeakerConfig dataclass
exceptions.py # Custom exceptions
client.py # BeakerClient (XML-RPC + REST)
models/ # Pydantic response models
servers/
__init__.py # FastMCP server, lifespan, DI helper
systems.py # System tools (4 read + 6 write)
jobs.py # Job tools (4 read + 6 write)
distros.py # Distro tools (2 read)
tasks.py # Task tools (1 read)
general.py # General tools (2 read)
prompts.py # Workflow prompt templates
resources.py # Beaker documentation resources
utils/
xml_validation.py # Job XML validation/auto-fill
diagnosis.py # Failure analysis engine
formatting.py # Human-readable formatters
bkr_cli.py # bkr CLI helpers
parsing.py # ID parsing utilities
Development
cd mcp-beaker
uv sync --dev
uv run pytest
uv run ruff check src/
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。