Skip to content

Wiggum as MCP server — expose core capabilities as MCP tools #95

@federiconeri

Description

@federiconeri

Summary

Expose Wiggum's project scanning, interview orchestration, and spec generation as MCP (Model Context Protocol) tools, making them available to any MCP-compatible agent or IDE (Claude Code, Cursor, Windsurf, VS Code, etc.).

Problem / Context

MCP is becoming the standard protocol for extending AI coding agents. By exposing Wiggum's unique capabilities as MCP tools, we make them available to every MCP-compatible environment without requiring users to install or switch to the Wiggum CLI.

Key architectural advantage: Competitors wrap agents (run on top of them). Wiggum can run inside agents via MCP — a fundamentally different integration model that's more natural and less disruptive to existing workflows. Combined with the skill.md distribution strategy, this creates a two-pronged distribution approach: skills for quick adoption, MCP for deep integration.

Roadmap phase: Phase 5 — DISTRIBUTE ("Wiggum meets users where they already are")

Proposed Solution

MCP Tools to Expose

Tool Description
wiggum_scan Run project scanner, return structured detection results
wiggum_analyze Run AI-enhanced codebase analysis
wiggum_interview_start Start an interactive interview session for a feature
wiggum_interview_answer Submit an answer to the current interview question
wiggum_interview_generate Generate spec from completed interview
wiggum_list_specs List available specs in the project
wiggum_read_spec Read a specific spec file

MCP Resources to Expose

Resource Description
wiggum://context Current .ralph/.context.json
wiggum://specs/{name} Individual spec files
wiggum://config Current ralph.config.cjs

Architecture

  • Standalone MCP server process (wiggum mcp-server or npx wiggum-mcp)
  • Uses stdio transport (standard for local MCP servers)
  • Reuses existing programmatic API (Scanner, Generator, ConversationManager)
  • Stateful session for multi-turn interview flow

Files to Modify

File Changes
src/mcp/ (new) MCP server implementation, tool handlers, resource providers
src/index.ts Add mcp-server command routing
package.json Add MCP server entry point, @modelcontextprotocol/sdk dependency

Acceptance Criteria

  • wiggum mcp-server starts a stdio-based MCP server
  • Scan and analyze tools return structured results usable by any agent
  • Interview flow works as multi-turn MCP tool calls
  • Spec resources are readable via MCP resource protocol
  • Installable in Claude Code via claude mcp add wiggum
  • Works in at least one additional MCP-compatible IDE (Cursor, Windsurf, or VS Code)
  • Integration tested with Claude Code end-to-end

Metadata

Metadata

Assignees

Labels

ai/llmAI workflows, agents, promptsfeatureNew capabilityintegrationsExternal APIs (Braintrust, Tavily, Context7, Sentry)

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions