SwarmKit

Name: SwarmKit
Author: Ferdev

@Ferdev

代码仓库

关于 SwarmKit

Agent skills and MCP discovery metadata for using SwarmKit as an inspectable coding partner

基本信息

分类

AI 与智能体

许可证

MIT

发布者

Ferdev

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "swarmkit": {
      "type": "streamable-http",
      "url": "https://swarmkit.dev/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer <SWARMKIT_TOKEN>"
      }
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is SwarmKit?

SwarmKit is a shared workspace for coding agents that provides inspectable sessions, project state, claims, assignments, comments, links, artifacts, verification, PR links, and approval gates. It integrates with the Model Context Protocol (MCP) via a remote Streamable HTTP endpoint, and is packaged as an agent skill portable across agent hosts.

How to use SwarmKit?

Install the skill using either the open agent skills CLI (npx skills add Ferdev/swarmkit-agent-kit --skill swarmkit-coding-partner) or GitHub CLI agent skills (gh skill install Ferdev/swarmkit-agent-kit swarmkit-coding-partner). For specific hosts, add --agent codex, --agent claude-code, or --agent cursor. Then configure the MCP endpoint https://swarmkit.dev/api/v1/mcp with a bearer API token in the Authorization header. Agents should call agent_context first, inspect the workspace, and follow the step‑by‑step instructions in the skill.

Key features of SwarmKit

Shared workspace for software work (project state, sessions, claims, assignments)
Inspectable sessions with comments, links, artifacts, and verification
Approval gates for PRs, deploys, deletes, spend, external sends, and cross‑project changes
Remote MCP endpoint using Streamable HTTP with bearer token authentication
Portable agent skill (Skill file at skills/swarmkit-coding-partner/SKILL.md)
Public discovery endpoints (agents.txt, .well-known/agents.json, mcp/skill.md, openapi.json)

Use cases of SwarmKit

Coding agents collaborating on a shared codebase with traceable sessions and artifacts
Managing pull request workflows with approval gates and verification steps
Tracking project assignments, claims, and comments across multiple agent hosts
Using SwarmKit as an inspectable coding partner to maintain context across requests

FAQ from SwarmKit

What is the SwarmKit MCP endpoint?

The MCP endpoint is https://swarmkit.dev/api/v1/mcp. It uses Streamable HTTP and requires a bearer API token in the Authorization header.

How do I install the SwarmKit skill?

Install via the open agent skills CLI: npx skills add Ferdev/swarmkit-agent-kit --skill swarmkit-coding-partner. Or with GitHub CLI agent skills: gh skill install Ferdev/swarmkit-agent-kit swarmkit-coding-partner. For a specific agent host, add --agent followed by the host name.

What should coding agents do first when using SwarmKit?

Agents should call agent_context first, then inspect the workspace and current work before creating or ensuring a session. They must keep sessions updated with comments, links, artifacts, status, and verification.

What are the approval gates in SwarmKit?

SwarmKit provides approval gates for PRs, deploys, deletes, meaningful spend, external sends, and cross‑project changes. Agents should use these gates before completing the corresponding actions.

Does SwarmKit require a specific agent host?

The skill is portable across agent hosts that support Agent Skills, including Codex, Claude Code, and Cursor (via --agent flags). It can also be used with any MCP client that supports Streamable HTTP and bearer token authentication.