Claude Desktop Integration
Claude Desktop can access the full Rebar knowledge base through the MCP filesystem server. This gives Claude Desktop the same context available in Claude Code — wiki pages, expertise.yaml, agent definitions, and client data.
How It Works
Claude Desktop supports MCP (Model Context Protocol) servers that provide file access. The @modelcontextprotocol/server-filesystem server gives Claude Desktop read access to specified directories.
The sync-obsidian.sh script already syncs framework files to a Windows-accessible path, so Claude Desktop can reach them without WSL path translation.
Setup
1. Sync Framework Files to Windows
The Obsidian sync script copies framework directories to a Windows path:
bash scripts/sync-obsidian.shThis syncs to a Windows-accessible location:
wiki/— knowledge wikiapps/— app configurations and expertiseclients/— client configurations and expertisesystem/— agent definitions and config.claude/commands/— slash command definitionsCLAUDE.md— framework instructions
2. Configure MCP in Claude Desktop
Add the filesystem server to your Claude Desktop MCP config (claude_desktop_config.json):
{
"mcpServers": {
"rebar": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"C:\\path\\to\\rebar\\wiki",
"C:\\path\\to\\rebar\\apps",
"C:\\path\\to\\rebar\\clients",
"C:\\path\\to\\rebar\\system"
]
}
}
}Restart Claude Desktop after editing the config.
3. Add Project Instructions
In Claude Desktop, create a project and add this to the project instructions:
Read wiki/index.md first for navigation. Follow [[wiki links]] to drill into topics.
Read system/paperclip.yaml for agent definitions.
Read clients/{name}/expertise.yaml for project-specific context.
This tells Claude Desktop where to start when answering questions about the framework.
What Claude Desktop Can Access
| Path | Content |
|---|---|
wiki/ | Knowledge wiki — patterns, decisions, platform knowledge |
wiki/index.md | Page summaries and navigation (read this first) |
apps/*/expertise.yaml | App-specific operational data |
clients/*/expertise.yaml | Client-specific operational data |
system/paperclip.yaml | Agent orchestration config |
system/agents/*.yaml | Detailed agent definitions |
Same Knowledge, Multiple Interfaces
The framework’s knowledge is stored in plain files. This means the same context is available across all interfaces:
| Interface | Access Method |
|---|---|
| Claude Code (CLI) | Direct file reads, slash commands |
| Claude Desktop | MCP filesystem server |
| Claude.ai (web) | Upload files or paste content |
| Obsidian | Direct vault on wiki/ directory |
| Quartz site | Rendered wiki at GitHub Pages URL |
The self-learn loop (expertise.yaml) and wiki grow through Claude Code sessions. Claude Desktop reads the results. The sync script keeps everything current.