Jev.aitools.fyi

Command Palette

Search for a command to run...

Sponsored

Cline and Roo Code + Jev

Last updated

MCP reflex tools in the VS Code sidebar agent. system1-mcp installer or typed evaluate servers.

What Cline and Roo are

Cline is an open VS Code extension agent with MCP server management in the Cline panel (stacked servers icon, Configure tab). Roo Code is a popular fork with a similar MCP story and compatible mcpServers JSON.

Both speak stdio MCP and can load community servers alongside built-in tools. Cline CLI reads ~/.cline/data/settings/cline_mcp_settings.json (see Cline configuration docs on GitHub).

Fastest path: system1-mcp installer

The system1-mcp PyPI package documents uvx system1-mcp install with autodetection for Cline, Roo Code, Cursor, Claude Desktop, Windsurf, and other hosts. It registers four reflex tools: fast_guard, fast_judge, fast_verify, and fast_score.

Keys resolve from TYPESAFE_API_KEY, ~/.system1/config.json, or workspace .env per the README.

uvx system1-mcp install
# Or non-interactive:
uvx system1-mcp install --api-key tsk_...
uvx system1-mcp doctor

Manual MCP block (Cline / Roo)

If you prefer hand-edited JSON, open Cline MCP Servers, Configure MCP Servers, and add an mcpServers entry. Roo uses the same top-level mcpServers shape in its MCP settings UI.

Prefer env interpolation (${env:TYPESAFE_API_KEY}) where supported instead of pasting secrets into committed files.

{
  "mcpServers": {
    "system1": {
      "command": "uvx",
      "args": ["system1-mcp"],
      "env": {
        "TYPESAFE_API_KEY": "${env:TYPESAFE_API_KEY}"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

typesafe-mcp evaluate (single evaluate tool)

itsmostafa/typesafe-mcp exposes evaluate for POST /v1/systemone shaped calls. Install via the project install.sh, then evaluate setup mcp when the README documents your client. Tool names differ from system1-mcp: you get a general evaluate surface rather than fast_guard presets.

curl -fsSL https://raw.githubusercontent.com/itsmostafa/typesafe-mcp/main/install.sh | sh
export TYPESAFE_API_KEY="tsk_..."
evaluate setup mcp

blakestone-x/jev-mcp (classify, score, check)

blakestone-x/jev-mcp documents Cursor-style uvx git installs and exposes jev-oriented tools (classify, score, check) for agents that want lower-level System One calls. Compare tool schemas before enabling two Jev servers at once, or the model will drown in duplicates.

Teach the agent to call reflex tools

MCP registration alone does not force usage. system1-mcp ships AGENTS.md guidance: call fast_guard before shell commands, fast_judge when choosing among candidates, fast_verify on logs.

Paste the snippet into your repo CLAUDE.md, AGENTS.md, or Cline custom instructions so the reflex layer runs on the hot path.

Environment keys

Create a key at https://console.typesafe.ai/ (or route through Vercel AI Gateway / OpenRouter where the integration documents it). Export TYPESAFE_API_KEY in the environment for Cline, Roo, and uvx MCP child processes.

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

Cline CLI MCP paths differ from the VS Code extension file; verify ~/.cline/data/settings/cline_mcp_settings.json for CLI runs. Restart or reconnect MCP after edits. fast_* tools are advisory: your harness still owns final execution.

Do not enable both system1-mcp and a second full Jev MCP server unless you need both tool shapes.

Related directory listings

  • Agent toolingGo62
    MCP
    typesafe-mcp
    Go CLI and single-binary MCP server exposing TypeSafe judgments to Claude Desktop, Claude Code, and Codex.

    by @0xLogicrw

    agent toolingmcpagents
    Details
  • Agent toolingPython7
    MCP
    jev-mcp (blakestone-x)
    MCP server for TypeSafe Jev: typed classify, score, check, match and screen for any agent, with confidence on every answer.

    by @0xLogicrw

    agent toolingmcpagents
    Details
  • Agent toolingTypeScript2
    MCPDemo
    jev-mcp (rashedInt32)
    MCP server exposing TypeSafe Jev as typed, calibrated judgment tools: classify, score, check, batched ask.
    agent toolingmcpagents
    DetailsDemo
  • Official205
    OfficialDemoFeatured
    TypeSafe agent skills
    Official Claude Code / Codex skill pack for Jev primitives, patterns, and evaluation shape.
    officialagentseval
    DetailsDemo

Related agent guides

FAQ

Does Roo use the same MCP file as Cline?
Roo documents its own MCP settings path in the Roo UI, but the JSON schema matches Claude-style mcpServers blocks. system1-mcp install targets both when autodetect succeeds.
fast_guard vs evaluate?
fast_guard is a preset shell safety tool in system1-mcp. evaluate is the generic System One caller in typesafe-mcp. Pick presets for IDE reflexes, evaluate when you design custom question maps in code.
Can Cline auto-approve Jev tools?
Cline MCP settings support autoApprove lists per server. Only auto-approve read-only tools you trust; keep fast_guard approval on if your policy requires human eyes on blocks.
Where are Cline MCP docs?
See Cline docs/mcp/mcp-overview.mdx on GitHub for Configure MCP Servers, remote streamableHttp servers, and CLI notes.
Roo-only features?
Roo adds product-specific modes and prompts, but Jev wiring is the same MCP contract. Re-run system1-mcp doctor after Roo updates if tools disappear.

Primary sources

All agent guidesJev primerAgent tooling category
By Rishit