MCP & Agent Tools

SootSim does not ship an MCP (Model Context Protocol) server, and that is a deliberate choice. The sootsim CLI is the agent integration surface, and it does more than an MCP wrapper around the same bridge could.

Every MCP-capable agent — Claude Code, Claude Desktop, Cursor, Codex — can already run shell commands. That means sootsim works in all of them today with zero MCP configuration.

#Why the CLI instead of MCP

An MCP server for SootSim would be a thin RPC layer over the same bridge the CLI already speaks. The CLI carries capabilities that layer would have to re-implement — and most of them only make sense as long-lived process state, which an MCP tool call does not have:

• Multi-agent isolation. sootsim claim takes an exclusive lease on a sim, so several agents can share one machine and one daemon without stealing each other’s taps. See sootsim claim.
• A persistent bridge daemon. The daemon keeps sims attached to their project by bundle URL across commands and across sessions. An MCP server would be one more process to supervise.
• Agent-aware auto-settling. Every command waits for the runtime to go idle before returning — no manual sleep, no screenshot-poll loops. The settle logic understands React Native’s commit/layout/animation phases.
• React Native-aware inspection. describe, find, what-happened, and diagnose report RN semantics — screens, keyboards, toasts, alerts, responder transfers — not just a DOM dump.
• A visible cursor and recordings. sootsim do animates a real cursor into the canvas and record captures it, so you can watch a remote or CI agent work.
• Composable in the shell. Pipe output, gate CI on sootsim assert exit codes, script flows — all without an extra protocol.

If a future agent host can only speak MCP and cannot run a shell, that is the moment to add a server. Until then, shipping one would mean two integration paths to keep in sync — and one path, no fallbacks, is the rule.

#Wiring an agent

There is nothing to configure. Make sure the CLI is installed and give the agent a sim to drive.

npm i -g sootsim
sootsim open 8081     # or your usual dev flow
sootsim list          # confirm a sim is connected

From there the agent runs sootsim commands directly. See Agents → Setup and Agents → CLI Use for the full workflow.

#Teach the agent with skills

The fastest way to get an agent productive is to drop in a SootSim skill — a markdown file the agent discovers automatically and uses to learn the CLI:

# generic — works in any agent-aware repo
npx skills add tamagui/soot --skill sootsim-debug

# or from a soot checkout
bun sootsim skills .claude/skills

Available: sootsim-debug, sootsim-a11y, sootsim-perf, sootsim-visual, sootsim-test-flow, sootsim-maestro, sootsim-detox. See the Agent Skills API and sootsim skills.

#What’s next

• Agents → Setup — install and verify the bridge.
• Agents → CLI Use — the commands an agent runs.
• Agent Devtool (Inspect) — the in-sim panel where you talk to an agent and watch it work.