MCP Servers (Beta)

Overview

Settings → MCP Servers (labeled Beta in the sidebar) lets you connect Model Context Protocol servers — external tool providers your AI can call during chat. An MCP server might expose your filesystem, a database, an issue tracker, or any other capability.

For how MCP tools actually behave in a conversation — tool approval, streaming results — see MCP Tools.

MCP servers are experimental and may not work as expected. They give the AI access to external tools which can read, modify, or transmit your data. Your conversations, prompts, and any data processed through MCP tools will be shared with the server. Only use MCP servers from sources you trust completely, and be extra careful with servers that can reach sensitive information.

---

Adding an MCP Server

1. Go to Settings → MCP Servers and click Add MCP Server.

2. Fill in the dialog:

   • Name (required) — how the server appears in the list
   • Description (optional)
   • Transport Type — how Slashspace talks to the server:
     • stdio (Local) — runs a local command to start the server. Provide the Command (e.g. npx -y @modelcontextprotocol/server-filesystem) and Arguments as a comma-separated list (e.g. a path the server is allowed to access).
     • HTTP (Remote) — connects to a remote server. Provide the URL, and optionally Headers (JSON), e.g. {"Authorization": "Bearer your-api-key"}.
     • SSE (Server-Sent Events) — like HTTP, with the same URL and Headers fields, for servers that stream over SSE.
   • Enabled — the switch at the bottom controls whether the server is active right away.

3. Click Add Server (or Update Server when editing).

---

Managing Servers

Each server is a row in an accordion:

• The plug icon is green when the server is enabled.
• A pill shows the transport type (stdio, http, or sse).
• The switch enables or disables the server without deleting it.
• Edit reopens the dialog with the server's settings.
• The trash button deletes the server after a confirmation — this cannot be undone and permanently removes the server configuration.

Expanding a row shows the description, the full transport command or URL, and the server's ID.

---

Use cases

• Filesystem access: run the filesystem MCP server over stdio so chat can read and summarize files in a folder you allow.
• Internal tooling: connect a remote HTTP MCP server your team hosts — with an Authorization header — to let the AI query internal systems.
• Trying community servers: enable a community MCP server for a one-off task, then flip its switch off so it can't be called until you need it again.
• Separating trust levels: keep powerful servers disabled by default and only enable the specific one a task needs, instead of leaving everything on.