Plugins, MCP connectors, and TypeSafe skills on Anthropic's agent stack.
What this stack is
Claude Code is Anthropic's terminal agent for software work. Claude Desktop includes Cowork, an agentic workspace that shares Claude Code's architecture for multi-step knowledge work on local files. ai, synced at session start.
claude directory unless you add the same capability in Customize.
Official TypeSafe agent skill
typesafe-ai/skills is the maintained skill for designing System One workflows. Claude Code can install it as a marketplace plugin; other agents use npx skills add.
claude plugin marketplace add typesafe-ai/skills
claude plugin install typesafe@typesafe-ai
# Other agents:
npx skills add typesafe-ai/skills --skill typesafe-aijev-judge-mcp (single judge tool)
gecm0/jev-judge-mcp registers one MCP tool, judge, for narrow typed questions over evidence you supply. Claude Code can install marketplace + plugin typesafe@jev in one step (server and skill bundled). Also supports generic stdio MCP config for Desktop.
claude plugin marketplace add gecm0/jev-judge-mcp
claude plugin install typesafe@jev
npx skills add gecm0/jev-judge-mcp --skill jev
npx skills add typesafe-ai/skills --skill typesafe-aitypesafe-mcp evaluate (Go MCP server)
itsmostafa/typesafe-mcp exposes an evaluate tool and can register with Claude Code, Claude Desktop, and Codex via evaluate setup mcp. It targets POST /v1/systemone (or OpenRouter decisions when configured).
curl -fsSL https://raw.githubusercontent.com/itsmostafa/typesafe-mcp/main/install.sh | sh
TYPESAFE_API_KEY=your-key evaluate setup mcpjevwire hooks (Brainwires)
Brainwires/jevwire ships MCP tools (jev_evaluate, jev_rank, jev_verify, jev_gate_action, and others) plus a Claude Code plugin whose hooks run at harness boundaries. Install via /plugin marketplace add Brainwires/jevwire and set TYPESAFE_API_KEY in plugin settings.
Cowork-specific notes
Cowork supports the same connectors and skills model as other Claude products: add MCP servers in Customize, enable skills, or install role plugins from Anthropic's knowledge-work-plugins repo.
Pair a Jev MCP connector with typesafe-ai/skills the same way Canva or Notion ship connector plus skill playbooks. Cowork changelog documents on-demand MCP tool loading (toolSearchEnabled) to reduce context from large tool schemas.
Environment keys
ai/ (or route through Vercel AI Gateway / OpenRouter where the integration documents it). Export TYPESAFE_API_KEY in the environment for Claude Code, Desktop, and Cowork sessions.
Some MCP clients accept AI_GATEWAY_API_KEY when using gateway-hosted Jev models; follow that integration's README. Never commit keys into repos or paste them into chat logs.
export TYPESAFE_API_KEY="tsk_..."
# Optional pin:
export TYPESAFE_MODEL="jev-latest"Pitfalls
Do not install jev-judge-mcp twice (plugin plus manual claude mcp add) or you will duplicate servers. Codex config may embed literal keys in TOML; prefer exporting TYPESAFE_API_KEY in the launching shell when possible. env files.
claude must be added in Customize to appear in Cowork sessions.
Related directory listings
Related agent guides
FAQ
- Does Cowork use Claude Code MCP config on disk?
- Cowork syncs enabled connectors and skills from Customize on your claude.ai account at session start, not from ~/.claude automatically.
- Which MCP should I pick?
- jev-judge-mcp if you want one opinionated judge tool and skill. typesafe-mcp if you want a single evaluate tool with multi-client setup script. jevwire if you want hook-based gates plus multiple tools.
- Can Desktop use the same plugin marketplace entries as Code?
- MCP connectors are supported across Claude products; exact UI paths differ. Follow each README for Claude Code vs Desktop registration.
- Where is Claude MCP documentation?
- See code.claude.com/docs for MCP quickstart and connector patterns.
- Does the TypeSafe skill call the API by itself?
- The skill teaches workflow design and points to docs. You still need an SDK, gateway, or MCP tool for live inference.