AI applications
SweepMe! can be controlled by AI agents via its built-in MCP server (from SweepMe! 1.6, part of SweepMe! Pro). This page helps you to choose an AI application and describes how to connect it to SweepMe!.
Choosing an AI application
Any AI application that supports the Model Context Protocol (MCP) with the "streamable HTTP" transport can be used, as long as it runs on the computer with SweepMe! or can reach it in the local network. AI services that connect to MCP servers from their own cloud, such as custom connectors in Claude or apps in ChatGPT on the web, cannot reach a lab computer.
Use what your organization provides
If your organization already provides an AI application that supports MCP, use it. Approved applications have usually been checked for data protection, licensing and IT security, and your IT department can support you with them. Examples of suitable applications are:
- Claude Desktop or Claude Code
- OpenAI Codex (CLI, IDE extension, or in the ChatGPT desktop app)
- GitHub Copilot in VS Code
- Cursor
- Google Antigravity
Some organizations restrict which MCP servers may be added to these applications. Ask your IT department if adding a local server is blocked.
Which data is shared: The language model receives everything that the SweepMe! tools return, e.g. measured values, parameter names, file paths and lines of the debug log. If the model runs in the cloud, this data leaves your computer. Check the rules of your organization regarding which data may be shared with an AI service.
Options for restricted IT environments
If no suitable application is available, or if cloud AI services are not allowed, the following applications are common choices. All of them can work with local language models, so that no data has to leave your network.
| Application | Type | Language models | Suited for |
|---|---|---|---|
| Msty Studio | Desktop application | Local models and cloud providers with your own API key | Individual users who want an easy to install desktop application |
| AnythingLLM | Desktop application, also available as server (Docker) | Built-in local models, local runtimes such as Ollama or LM Studio, and cloud providers | Individual users, fully offline setups |
| LibreChat | Web application that is hosted by your organization (usually Docker) | The providers that the administrator configures, e.g. company accounts of cloud providers or local models | Teams, if IT wants to manage users and models centrally |
LM Studio is another desktop application to run local models and can connect to SweepMe! directly.
Please consider:
- Tool support: The language model must support tool calling. Small local models often use tools less reliably than large cloud models. Larger local models need a powerful computer with a graphics card.
- Network: The SweepMe! MCP server has no authentication. If the AI application runs on another computer, e.g. a central LibreChat server, it must be able to reach the lab computer, and the MCP server must listen on a network address. Only do this in a trusted network, or run the AI application on the lab computer itself.
- Approval: Even open source applications usually need the approval of your IT department before they can be installed.
Connecting AI applications
First start the MCP server in SweepMe! as described in Interfaces. The examples use the server name sweepme and the default URL http://localhost:8082/mcp. Replace the URL if you changed host or port. Most applications must be restarted, or the server list refreshed, after the configuration was changed. Any other MCP client that supports streamable HTTP can be connected with the same URL. Clients that only support local (stdio) servers can use the bridge mcp-remote as shown for Claude Desktop. If the connection does not work, see Troubleshooting MCP connections.
Claude Desktop
Claude Desktop connects to local servers only via its configuration file, using the bridge mcp-remote. This requires Node.js. Open Settings → Developer → Edit Config (file %APPDATA%\Claude\claude_desktop_config.json) and add:
{
"mcpServers": {
"sweepme": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://localhost:8082/mcp"]
}
}
}
Quit Claude Desktop completely and start it again. Custom connectors (Settings → Connectors) do not work for SweepMe!, because Claude connects to them from Anthropic's cloud.
Claude Code
claude mcp add --transport http --scope user sweepme http://localhost:8082/mcp
Without --scope user, the server is only added to the current project.
OpenAI Codex
The Codex CLI, the Codex IDE extension and Codex in the ChatGPT desktop app share the same configuration. Either run
codex mcp add sweepme --url http://localhost:8082/mcp
or add the following to %USERPROFILE%\.codex\config.toml:
[mcp_servers.sweepme]
url = "http://localhost:8082/mcp"
In the app, the server can also be added via Settings → MCP servers → Add server with the type Streamable HTTP.
Google Antigravity
In the agent panel, open … → MCP Servers → Manage MCP Servers → View raw config and add:
{
"mcpServers": {
"sweepme": {
"serverUrl": "http://localhost:8082/mcp"
}
}
}
Save the file and refresh the list of installed MCP servers. The Antigravity CLI uses the same configuration.
VS Code (GitHub Copilot)
Run the command MCP: Add Server from the Command Palette, select HTTP and enter the URL. Alternatively, add the server to the file .vscode/mcp.json of your workspace:
{
"servers": {
"sweepme": {
"type": "http",
"url": "http://localhost:8082/mcp"
}
}
}
Confirm the trust prompt when VS Code starts the server.
Cursor
Add to %USERPROFILE%\.cursor\mcp.json (all projects) or .cursor/mcp.json (one project):
{
"mcpServers": {
"sweepme": {
"url": "http://localhost:8082/mcp"
}
}
}
Then switch the server on in the MCP settings of Cursor.
Website: https://cursor.com/
Windsurf / Devin Desktop
Open the MCP configuration file from the … menu of the agent panel (Open MCP config file) and add:
{
"mcpServers": {
"sweepme": {
"serverUrl": "http://localhost:8082/mcp"
}
}
}
Website: https://devin.ai/desktop
Msty Studio
Open the Toolbox, select Add New Tool → HTTP, enter a name and the MCP server URL http://localhost:8082/mcp, and click Add. Check the connection in the Tool Console.
More information: Tools in the Msty Studio documentation
Website: https://msty.ai/products/studio/
LM Studio
LM Studio (0.3.17 or later) can use SweepMe! together with a local language model, e.g. on a computer without internet access. Open the Program tab in the right sidebar, select Install → Edit mcp.json and add:
{
"mcpServers": {
"sweepme": {
"url": "http://localhost:8082/mcp"
}
}
}
Website: https://lmstudio.ai/download
AnythingLLM
AnythingLLM can also be used with a local language model. Add the server to the file plugins\anythingllm_mcp_servers.json in the storage folder of AnythingLLM (Desktop on Windows: %APPDATA%\anythingllm-desktop\storage). The file is created when you open the Agent Skills page for the first time.
{
"mcpServers": {
"sweepme": {
"type": "streamable",
"url": "http://localhost:8082/mcp"
}
}
}
Then click Refresh on the Agent Skills page to load the server. MCP tools are used by the agent, so start a chat with @agent.
More information: MCP on AnythingLLM Desktop, MCP compatibility overview (configuration format)
Website: https://anythingllm.com/
LibreChat
Add the server to librechat.yaml and restart LibreChat. LibreChat blocks connections to local addresses by default, so the address must be allowed explicitly:
mcpServers:
sweepme:
type: streamable-http
url: http://localhost:8082/mcp
mcpSettings:
allowedAddresses:
- "localhost:8082"
If LibreChat runs in Docker, use host.docker.internal instead of localhost in both places. In this case, the container must be able to reach the host and port that are set for the MCP server in SweepMe!.
More information: MCP in the LibreChat documentation, MCP settings (Click to see how to setup allowedAddresses)
Website: https://www.librechat.ai/