swarm-kb 0.4.0

Shared knowledge base for the Swarm suite (ReviewSwarm, FixSwarm, DocSwarm, ArchSwarm, SpecSwarm)

swarm-kb

Shared knowledge base for the Swarm AI suite: ReviewSwarm, FixSwarm, DocSwarm, ArchSwarm.

Provides centralized session storage, cross-tool references, and code mapping so all swarm tools share a single knowledge layer at ~/.swarm-kb/.

Installation

pip install swarm-kb

MCP Server Setup

Claude Code

claude mcp add swarm-kb -- swarm-kb serve --transport stdio

Cursor / Windsurf / Cline (SSE)

swarm-kb serve --port 8788

Then add to your MCP config:

{
  "mcpServers": {
    "swarm-kb": {
      "url": "http://localhost:8788/sse"
    }
  }
}

Manual .mcp.json (per-project)

{
  "mcpServers": {
    "swarm-kb": {
      "type": "stdio",
      "command": "swarm-kb",
      "args": ["serve", "--transport", "stdio"]
    }
  }
}

CLI

swarm-kb status          # Show KB health: session counts, storage root
swarm-kb serve           # Start MCP server (SSE on port 8788)
swarm-kb serve --transport stdio  # Start MCP server (stdio for Claude Code)

Storage Layout

~/.swarm-kb/
├── sessions/
│   ├── review/     # ReviewSwarm sessions
│   ├── doc/        # DocSwarm sessions
│   ├── arch/       # ArchSwarm sessions
│   └── fix/        # FixSwarm sessions
├── code-maps/      # AST-based code analysis per project
└── xrefs.jsonl     # Cross-tool references

Migration

On first startup, swarm-kb automatically migrates data from legacy tool-specific directories:

• ~/.review-swarm/ -> ~/.swarm-kb/sessions/review/
• ~/.doc-swarm/ -> ~/.swarm-kb/sessions/doc/
• .archswarm_sessions/ -> ~/.swarm-kb/sessions/arch/

Suite Overview

Tool	Package	Purpose
swarm-kb	swarm-kb	Shared knowledge base & code maps
ReviewSwarm	review-swarm	Collaborative AI code review
DocSwarm	doc-swarm-ai	Documentation generation
FixSwarm	fix-swarm-ai	Automated fix planning & application
ArchSwarm	arch-swarm-ai	Architecture analysis & debate

Install the full suite:

pip install swarm-kb review-swarm doc-swarm-ai fix-swarm-ai arch-swarm-ai

License

MIT