Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@interactive-cli Use interactive-cli to run 'eas build --platform ios' and choose the production profile."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP server that lets AI coding agents interact with interactive CLI processes. Uses a real PTY + xterm headless to render terminal output as clean text — no ANSI escape soup.
Works with Claude Code, Cursor, Windsurf, or any MCP-compatible client.
AI agents can run shell commands, but they choke on anything interactive:
$ eas build --platform ios
✔ Select a build profile › (waiting for input...)This MCP server bridges that gap — spawn the process, read the screen, send keystrokes.
Related MCP server: pty-mcp
claude mcp add-json interactive-cli '{"command":"npx","args":["-y","interactive-cli-mcp"]}'Or add to ~/.claude/settings.json manually:
{
"mcpServers": {
"interactive-cli": {
"command": "npx",
"args": ["-y", "interactive-cli-mcp"]
}
}
}git clone https://github.com/ohernandezdev/interactive-cli.git
cd interactive-cli
npm install && npm run buildclaude mcp add-json interactive-cli '{"command":"node","args":["/absolute/path/to/interactive-cli/dist/index.js"]}'Point your client's MCP config at:
command: node
args: ["/path/to/interactive-cli/dist/index.js"]
transport: stdioTool | Description | Read-only |
| Start an interactive process in a PTY | No |
| Type text (appends Enter by default) | No |
| Send special keys in sequence (arrows, ctrl combos, etc.) | No |
| Capture current terminal screen, optional regex search | Yes |
| Block until a regex pattern appears in output | No |
| Block until the process exits | No |
| Change terminal dimensions | No |
| List all active sessions | Yes |
| Kill process, clean up session | No |
sequenceDiagram
participant Agent as AI Agent
participant MCP as interactive-cli
participant PTY as Process (PTY)
Agent->>MCP: spawn("eas build --platform ios")
MCP->>PTY: Start process in PTY
PTY-->>MCP: Initial output
MCP-->>Agent: sessionId + screen
Agent->>MCP: get_screen(sessionId)
MCP-->>Agent: "Select profile: ❯ dev staging prod"
Agent->>MCP: send_keys(["down","down","enter"])
MCP->>PTY: ↓ ↓ ⏎
PTY-->>MCP: "✔ Selected: production"
MCP-->>Agent: updated screen
Agent->>MCP: wait_for("Build complete|error", 600s)
loop Every 1s
MCP->>PTY: check output
end
PTY-->>MCP: "Build complete"
MCP-->>Agent: matched + screen
Agent->>MCP: close(sessionId)
MCP->>PTY: SIGTERM
MCP-->>Agent: final screen + exit codespawnspawn({
command: "ssh user@server.com", // full command string
cwd: "/path/to/project", // optional working directory
env: { "NODE_ENV": "production" }, // optional extra env vars
cols: 120, // terminal width (default 120)
rows: 30, // terminal height (default 30)
waitMs: 3000, // ms to wait for initial output (default 2000)
})Commands run through your shell ($SHELL or /bin/zsh), so pipes, redirects, and builtins work.
send_inputsend_input({
sessionId: "s1",
text: "yes", // text to type
pressEnter: true, // append Enter (default true). false for password fields
waitMs: 2000, // ms to wait for response (default 2000)
})send_keysSend one or more keys in sequence with a small delay between each:
send_keys({
sessionId: "s1",
keys: ["down", "down", "enter"], // navigate menu
delayBetweenMs: 50, // ms between keys (default 50)
waitMs: 1500, // ms to wait after last key (default 1500)
})Supported keys: enter tab escape space up down left right backspace delete home end page_up page_down f1–f12 ctrl+c ctrl+d ctrl+z ctrl+l ctrl+a ctrl+e ctrl+r ctrl+w ctrl+u ctrl+k ctrl+p ctrl+n y n 0–9
get_screenget_screen({
sessionId: "s1",
search: "error|warning", // optional regex to highlight matching lines
})
// Returns: { screen, cursor: { row, col }, searchResults, stats }wait_forwait_for({
sessionId: "s1",
pattern: "\\$|#|>", // regex to match (case-insensitive)
timeoutMs: 30000, // max wait (default 30s)
intervalMs: 1000, // check interval (default 1s)
})
// Returns: { matched: true/false, screen, elapsed }wait_for_exitwait_for_exit({
sessionId: "s1",
timeoutMs: 60000,
})
// Returns: { alive: false, exitCode: 0, screen }Sessions are exposed as MCP resources:
interactive-cli://sessions — JSON list of all sessions
interactive-cli://sessions/{id}/screen — live screen content
Pre-built prompt templates for common flows (appear as slash commands in Claude Code):
eas_build — EAS build with credential handling
ssh_session — SSH connection with command execution
repl_session — Start a REPL (Python, Node, psql, etc.)
docker_interactive — Run an interactive Docker container
graph LR
A[AI Agent] <-->|MCP Protocol| B[interactive-cli server]
B --> C[Session Manager]
C --> D[node-pty<br/><i>real PTY</i>]
C --> E[xterm headless<br/><i>screen render</i>]
C --> F[Truncation<br/><i>80K limit</i>]
D <--> G[Child Process<br/>ssh, eas, python...]node-pty spawns a real pseudo-terminal, so the child process thinks it's talking to a human
xterm headless maintains a virtual terminal buffer that renders ANSI sequences into a clean 2D text grid
Output is truncated at 80K chars (MCP clients like Claude Code cap at 100K) with smart middle-truncation preserving start and end
Tool annotations (readOnlyHint, destructiveHint) tell the client which tools are safe to run in parallel and which need permission
Node.js >= 18
macOS or Linux (node-pty uses native PTY)
MIT
This server cannot be deployed
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
Develop, manage, and debug Railway projects, services, and deployments from within agents.
Control real Android and iOS devices with LLM agents — tap, swipe, type, automate flows.
Let AI operate servers without SSH. Choose actions, approve risky changes, and audit every step.
