octopilot-mcp
Model Context Protocol (MCP) server for Octopilot — enables AI agents to detect, generate, build, and wire up new repositories end-to-end using the Octopilot CI/CD toolchain.
README
octopilot-mcp
Model Context Protocol (MCP) server for Octopilot — enables AI agents to detect, generate, build, and wire up new repositories end-to-end using the Octopilot CI/CD toolchain.
What it does
| Tool | Description |
|---|---|
detect_project_contexts |
Parse skaffold.yaml → pipeline-context JSON (languages, versions, matrix) |
generate_skaffold_yaml |
Generate a skaffold.yaml for given artifacts |
generate_ci_workflow |
Full .github/workflows/ci.yml using the standard octopilot pipeline |
onboard_repository |
One-call onboarding: detect → generate files → return next steps |
run_op_build |
Run op build via local binary or ghcr.io/octopilot/op container |
list_actions |
All Octopilot GitHub Actions from the bundled registry |
get_action_details |
Full spec, inputs, examples, gotchas for one action |
op promote-imageis intentionally not exposed. Image promotion between environments is operationally sensitive and must only run through a GitHub Actions workflow (with audit trail, OIDC credentials, and environment protection rules). Usegenerate_ci_workflowto produce the workflow that handles promotion safely.
Option A — Hosted (zero install)
Connect directly to the public server at https://mcp.octopilot.app — no cloning, no Python, no pip required.
# Cursor
fastmcp install cursor https://mcp.octopilot.app --name octopilot
# Claude Desktop
fastmcp install claude https://mcp.octopilot.app --name octopilot
Available hosted tools (stateless, no local dependencies):
| Tool | Description |
|---|---|
list_actions |
Browse the Octopilot GitHub Actions registry |
get_action_details |
Full spec, inputs, examples, gotchas for one action |
generate_skaffold_yaml |
Generate a skaffold.yaml for given artifacts |
generate_ci_workflow |
Full .github/workflows/ci.yml for a project |
Need
detect_project_contexts,onboard_repository, orrun_op_build? Those tools need Docker and local filesystem access — use Option B below.
Option B — Local install (full suite)
# Clone and install
git clone https://github.com/octopilot/octopilot-mcp
cd octopilot-mcp
uv sync
Usage
Register with your IDE (one command, FastMCP 3 CLI)
Docker or Colima is the only external dependency. Most tools are pure Python;
run_op_build pulls ghcr.io/octopilot/op:latest automatically with
--pull always, so you always run the latest release.
# Cursor
uv run fastmcp install cursor src/octopilot_mcp/server.py --name octopilot
# Claude Desktop
uv run fastmcp install claude src/octopilot_mcp/server.py --name octopilot
Development (hot-reload)
uv run fastmcp dev src/octopilot_mcp/server.py --reload
Inspect tools from the terminal
# List all available tools
uv run fastmcp list src/octopilot_mcp/server.py
# Call a tool directly
uv run fastmcp call src/octopilot_mcp/server.py tool_list_actions
Run as a server directly
uv run octopilot-mcp
Manual JSON config (alternative to fastmcp install)
{
"mcpServers": {
"octopilot": {
"command": "uv",
"args": ["run", "--directory", "/path/to/octopilot-mcp", "octopilot-mcp"]
}
}
}
Pin to a specific op release (optional):
{
"mcpServers": {
"octopilot": {
"command": "uv",
"args": ["run", "--directory", "/path/to/octopilot-mcp", "octopilot-mcp"],
"env": { "OP_IMAGE": "ghcr.io/octopilot/op:v1.0.0" }
}
}
}
Environment variables
| Variable | Default | Description |
|---|---|---|
OP_IMAGE |
ghcr.io/octopilot/op:latest |
Pin to a specific op release for reproducibility |
Example agent interaction
User: Onboard this Rust API project to use Octopilot CI.
Agent: [calls onboard_repository("/path/to/my-api", "ghcr.io/my-org")]
→ Detected: rust (stable) in api/
→ Generated: skaffold.yaml, .github/workflows/ci.yml
→ Next steps: add .pre-commit-config.yaml, push changes
Development
uv sync --all-extras
# Run tests
uv run pytest tests/ -v
# Run with coverage
uv run pytest tests/ --cov=src/octopilot_mcp --cov-report=term-missing
Tool module coverage target: ≥95% (actions, detect, generate, op_runner).
See CONTRIBUTING.md for details.
Resources
The server also exposes MCP resources for agent context:
octopilot://actions— Full actions registry JSONoctopilot://pipeline-context-schema— JSON Schema for pipeline-contextoctopilot://docs/getting-started— Plain-text onboarding guideoctopilot://docs/skaffold-patterns— Commonskaffold.yamlpatterns
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。