Give AI Agents Screen Control: OverRec's MCP Server

AI coding agents are good at reasoning about a task, but bad at touching the screen — finding a window, lining it up, and grabbing a clean screenshot has always meant gluing together ad-hoc scripts. OverRec closes that gap by running as a Model Context Protocol (MCP) server, so any MCP-aware agent — Claude Code, Claude Desktop, or your own tooling — can call OverRec’s window and screen tools directly, with no custom glue code.

---

What the MCP Server Exposes

Every OverRec CLI command is mirrored as an MCP tool:

Tool	Mirrors	Notes
monitors	cli monitors	Full details by default, or a compact ID/resolution list
window	cli window	Find windows by keyword
screenshot	cli screenshot	Capture a region or a specific window by ID
draw	cli draw	Show a click-through overlay rectangle to verify an area
record	cli record	Capture a region to GIF or MP4, in the background or blocking
snap	cli snap	Move and resize a window to exact coordinates
version	cli version

An agent can chain these in one conversation: list monitors, find a window, snap it into place, draw a confirmation rectangle, then capture the final screenshot — all as tool calls, no shell scripting required.

---

Starting the Server

OverRec supports two transports:

# Streamable HTTP — start once, share across clients
OverRec.exe cli mcp

# stdio — the client launches and manages the process itself
OverRec.exe cli mcp --stdio

HTTP mode listens on 127.0.0.1:7820 by default and serves a stateless JSON-RPC endpoint at /mcp; non-local Origin headers are rejected to prevent DNS rebinding. stdio mode speaks newline-delimited JSON-RPC over stdin/stdout and exits when the client closes the pipe — the simpler choice when the agent itself is responsible for starting OverRec.

---

Registering with Claude Code

# Option A — HTTP, leave the server running
OverRec.exe cli mcp
claude mcp add --transport http overrec http://127.0.0.1:7820/mcp

# Option B — stdio, Claude Code manages the process
claude mcp add overrec -- OverRec.exe cli mcp --stdio

Once registered, Claude can call monitors, window, screenshot, draw, record, and snap like any other tool — describing the screen action it wants in plain language instead of issuing raw CLI commands.

---

Inspecting the Server Yourself

The screenshot above is the MCP Inspector connected to OverRec over stdio. It lists every tool with its description, lets you run one manually, and shows the raw JSON-RPC result — useful for confirming the server works before wiring it into an agent, or for debugging a tool call that didn’t do what you expected.

---

Why This Matters

Screenshots and window control are exactly the kind of task that’s brittle when scripted ad hoc — paths differ per machine, window titles vary, coordinates drift. MCP gives agents a stable, typed interface to OverRec instead: the agent asks for “the Chrome window at 800x600” and gets back a window ID, a snapped position, and a screenshot, the same way every time.

That makes OverRec’s MCP server useful for:

• agent-driven documentation and bug-report screenshots
• automated UI comparison across runs
• any workflow where an AI agent needs to see and control the screen, not just reason about code

Download OverRec from the Microsoft Store.