Warp AI Setup

Warp supports MCP servers and AI rules but has no file-based global config — it mixes UI settings with file detection. This skill documents the exact toggles and files so our Claude tooling propagates to Warp AI.

One-time UI toggles

1. Settings → AI → MCP Servers → "File-based MCP Servers" — enable
   • Effect: Warp reads ~/.claude.json (Claude Code format) and any project-root .mcp.json
2. Settings → AI → Agents — verify default agent is set (otherwise /agent commands fail)
3. Warp Drive → Personal → Rules → Global — optional: paste core identity rules (UI-only; no file mirror). Keep short; detailed rules live in project AGENTS.md.

File-based rules (per-project)

Warp auto-detects any of these at project root:

• AGENTS.md (Warp-preferred)
• WARP.md (legacy)
• Symlinks to CLAUDE.md, .cursorrules, .clinerules, .github/copilot-instructions.md

Fast setup in a repo:

ln -s CLAUDE.md AGENTS.md   # if you have a Claude-native file
# or
cp ~/.claude/CLAUDE.md ./AGENTS.md

MCP servers

After enabling the toggle above, Warp reads ~/.claude.json which the sync-ai-tools skill keeps up to date. To add a new server:

1. Add to ~/.cursor/mcp.json (source of truth)
2. Run bash ~/.claude/skills/sync-ai-tools/sync.sh
3. Restart Warp (or it may auto-reload)

What Warp does NOT support

• No file-based skill system (Warp has no equivalent to Claude's ~/.claude/skills/)
• No file-based global rules (must use UI)
• No PreToolUse/PostToolUse hooks

Fallbacks

• For "skill-like" extension, use MCP servers (agents call MCP tools).
• For "rules-like" persistence, use project AGENTS.md (commit to repo) or Warp Drive Rules (personal).
• For workflows, continue using ~/.warp/workflows/*.yaml.

Verify

# Check Warp can read Claude's MCP set
jq '.mcpServers | keys' ~/.claude.json

Should list: fetch, filesystem, playwright, context7, memory, github, sequential-thinking.