Initializing SDK Environment...

MCP SERVER

RouteRTL MCP Server

The RouteRTL MCP server exposes CLI knowledge to AI agents via the Model Context Protocol. Any MCP-capable tool (Claude Code, Cursor, Windsurf, etc.) gains structured access to project configuration, the IP registry, FPGA knowledge base, design rules, and diagnostic guidance.

Installation

The MCP server ships inside the routertl package as an optional extra:


pip install routertl[mcp]

For development installs:


pip install -e ".[mcp]"

This installs the routertl-mcp command.

Configuration

Claude Code

Add to your project's .mcp.json (recommended) or ~/.claude/.mcp.json:


{
  "mcpServers": {
    "routertl": {
      "command": "routertl-mcp"
    }
  }
}

Restart Claude Code. The 12 rr_* tools will appear automatically.

Cursor / Windsurf / Other MCP Clients

Consult your editor's MCP documentation. The server runs on stdio transport by default:


routertl-mcp          # stdio (default)
routertl-mcp --sse    # SSE on port 8750

Available Tools

Tool	Description
`rr_help`	Structured CLI command reference (49 commands + 13 pkg subcommands)
`rr_project_info`	Parse and return the resolved project.yml configuration
`rr_list_targets`	Discover all available build targets
`rr_pkg_search`	Search the IP package registry
`rr_pkg_info`	Detailed metadata for a specific package
`rr_diagnose`	Pattern-match errors against known fixes + live doctor check
`rr_rules_check`	Run FPGA design rules evaluation
`rr_rules_list`	List available design rules
`rr_ask`	Query the FPGA knowledge base (fpgawisdom.com)
`rr_doctor`	Environment health check
`rr_status`	Quick project status snapshot
`rr_deps`	Show the dependency graph

Resources

The server also exposes two MCP resources:

routertl://commands — complete CLI command reference
routertl://project — current project.yml as YAML

Architecture

The server uses a hybrid approach:

Python API imports for structured queries (rr_project_info, rr_pkg_search, rr_ask, rr_list_targets) — fast, no subprocess overhead, returns structured data.
CLI wrappers for everything else (rr_doctor, rr_status, rr_rules_check, etc.) — reuses existing CLI formatting and output.
Static knowledge for rr_help and rr_diagnose — no I/O needed for basic command discovery and common error patterns.

Verification

After installation, verify the server loads correctly:


routertl-mcp --help

# Or test the module directly:
python -c "from sdk.mcp_server import mcp; print(f'{len(mcp._tool_manager._tools)} tools registered')"

Loading docs...