EVOKORE-MCP

# EVOKORE-MCP

EVOKORE-MCP is a TypeScript-based stdio MCP router and multi-server aggregator. It gives AI clients a single MCP endpoint that combines EVOKORE’s native workflow tools with proxied child servers defined in `mcp.config.json`, while adding namespace isolation, dynamic tool discovery, and human-in-the-loop approval controls.

Current package/runtime version: `3.0.0`.

## Current operator snapshot

- **Runtime shape:** stdio MCP router plus multi-server child aggregation
- **Native tool surface:** `docs_architect`, `skill_creator`, `resolve_workflow`, `search_skills`, `get_skill_help`, `discover_tools`, `proxy_server_status`, `refresh_skills`, `fetch_skill`, `execute_skill`, `list_registry`
- **v3.0 capabilities:** RBAC permissions, rate limiting, HTTP transport, MCP resources/prompts, tool annotations, skill ecosystem (versioning, registry, sandbox), session dashboard, async proxy boot
- **Continuity surfaces:** canonical session manifest, managed Cursor memory sync, manifest-backed status line, repo-state audit CLI
- **Voice surfaces:** proxied ElevenLabs tools, VoiceMode guidance, standalone VoiceSidecar with persona-aware hook forwarding
- **Recent change report:** [docs/RECENT_ADDITIONS_2026-03-12.md](docs/RECENT_ADDITIONS_2026-03-12.md)

## Why EVOKORE exists

EVOKORE exists to solve three common MCP operator problems:

- **Too many disconnected MCP endpoints**: EVOKORE collapses multiple child servers into one stdio server.
- **Too much tool-context overhead**: EVOKORE can run in `legacy` mode for broad compatibility or `dynamic` mode for session-scoped tool activation.
- **Too little operational control**: EVOKORE adds HITL approval, duplicate-prefix collision handling, runtime guardrails, and continuity docs for long-running repo work.

## Current capabilities

- **Single stdio MCP endpoint** for native EVOKORE tools plus proxied child servers.
- **11 native tools**: `docs_architect`, `skill_creator`, `resolve_workflow`, `search_skills`, `get_ski