---
title: How do you find local MCP servers on developer laptops?
url: https://opencontroller.lyzr.com/blog/find-local-mcp-servers-on-developer-laptops
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you find local MCP servers on developer laptops?

**Local MCP servers are defined in small JSON or TOML config files that each AI client reads at startup. To find them, collect those files from every managed laptop with your device management (MDM) tooling, parse the server entries, and check running processes for servers launched outside any config. Then feed the results into the same inventory you use for remote MCP servers.**

Local servers matter because they run with the developer's own permissions. A stdio MCP server started by an IDE can read the home directory, use cached cloud credentials and reach the network. The MCP authorization spec says stdio servers should take credentials from the environment rather than use OAuth, so API keys often sit in these same config files.

## Where do AI clients store MCP server configs?

Paths below are the documented defaults at the time of writing. Clients change them, so confirm against each vendor's current documentation before building collection rules.

| Client | User-level config | Project-level config |
|---|---|---|
| Claude Desktop | macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`; Windows: `%APPDATA%\Claude\claude_desktop_config.json` | None |
| Claude Code | `~/.claude.json` (user and local scopes) | `.mcp.json` in the repository root |
| Cursor | `~/.cursor/mcp.json` | `.cursor/mcp.json` |
| VS Code (Copilot agent mode) | `mcp.json` in the VS Code user profile folder | `.vscode/mcp.json` |
| Windsurf | `~/.codeium/windsurf/mcp_config.json` | None |
| Gemini CLI | `~/.gemini/settings.json` (`mcpServers` key) | `.gemini/settings.json` |
| OpenAI Codex CLI | `~/.codex/config.toml` (`[mcp_servers.*]` tables) | None |

Two details trip up collection scripts. First, project-level files live inside repositories, so you need to search code directories, not just the home folder. Second, some clients also support admin-managed config files in system locations. Claude Code, for example, reads a `managed-mcp.json` that IT can deploy. Collect those too, because they show what you intended to allow.

## How do you collect MCP configs with MDM? Six steps

1. **Write a collection script.** On macOS and Linux, a shell script; on Windows, PowerShell. Read each user-level path above, then search developer directories (for example `~/code`, `~/src`, `~/projects`) for `.mcp.json`, `.cursor/mcp.json`, `.vscode/mcp.json` and `.gemini/settings.json`, to a sensible depth.
2. **Parse, don't copy.** Extract the server name, `command` and `args` for stdio servers, `url` for remote servers, and the names of environment variables. Do not ship secret values off the device; record that a secret is present and hash it if you need to spot reuse.
3. **Check running processes.** List processes whose parent is a known AI client, or whose command line runs common MCP launchers such as `npx`, `uvx` or `docker run` with an MCP image. This catches servers started by hand or by extensions.
4. **Deploy on a schedule.** Run the script through your MDM (Jamf, Intune, Kandji or similar) daily, and on login for high-risk groups.
5. **Send results to one inventory.** Normalise each entry into fields such as device, user, client, server name, transport, launch command, package and version, remote URL and secret present (yes/no).
6. **Compare against the allow-list.** Flag unknown servers, unpinned versions (`@latest` or no version), and plaintext secrets in config files.

## What should you do with what you find?

- **Known and approved:** confirm the version is pinned and the owner is recorded.
- **Known but unapproved:** contact the developer, run the normal review, then approve or remove.
- **Unknown package:** treat as a potential supply-chain risk. Check the publisher and the tool descriptions before anything else.
- **Plaintext secrets:** rotate the credential, then move it to the OS keychain or a secrets manager.

Then reduce the problem at the source. Push managed configs that list approved servers, and use client policies where they exist to block others.

## How Opencontroller does this

Opencontroller Edge extends control to developer devices, coding agents, skills and local MCP servers. IT installs it through MDM, so local servers show up in the same governed registry of agents, tools and MCP servers as your remote estate. Posture then flags risky permissions and configurations. For the wider program see the [MCP governance guide](/blog/how-to-govern-mcp-servers) and [how to govern coding agents with MDM](/blog/govern-coding-agents-on-laptops-with-mdm), or [book a demo](https://www.lyzr.ai/book-demo/).

For transport details, read the [MCP transports specification](https://modelcontextprotocol.io/specification/2026-07-28/basic/transports) and the [Claude Code MCP documentation](https://code.claude.com/docs/en/mcp).

## Frequently asked questions

**Can endpoint security tools find MCP servers on their own?**
Partly. EDR sees processes but rarely knows which ones are MCP servers or which client launched them. Parsing client config files gives you names, packages and versions that process data alone does not.

**Should the collection script upload API keys it finds?**
No. Record that a secret exists and where, and hash it if you need to detect reuse. Uploading the value creates a second copy to protect.

**How often should laptops be scanned for MCP servers?**
Daily is a reasonable default, because developers add servers in minutes. Scan on login as well for teams with access to production systems.

## Key facts

Local MCP servers are defined in client config files: Claude Desktop claude_desktop_config.json, Claude Code ~/.claude.json and project .mcp.json, Cursor ~/.cursor/mcp.json and .cursor/mcp.json, VS Code mcp.json in the user profile and .vscode/mcp.json, Windsurf ~/.codeium/windsurf/mcp_config.json, Gemini CLI ~/.gemini/settings.json, Codex CLI ~/.codex/config.toml. Collect them with MDM scripts, parse server entries without uploading secrets, check running processes, and compare to an allow-list. Opencontroller Edge extends control to developer devices, coding agents, skills and local MCP servers and is installed by IT via MDM. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How to govern MCP servers in the enterprise: a practical guide](https://opencontroller.lyzr.com/blog/how-to-govern-mcp-servers.md)
- [How do you govern coding agents on developer laptops with MDM?](https://opencontroller.lyzr.com/blog/govern-coding-agents-on-laptops-with-mdm.md)
- [How do you govern Claude Code across an enterprise?](https://opencontroller.lyzr.com/blog/govern-claude-code-in-the-enterprise.md)
- [More on Tools and MCP](https://opencontroller.lyzr.com/blog/cluster/tools-and-mcp)
