Home / Features / Realtime / MCP server fleet
Hosted MCP server fleet on osFoundry
osFoundry hosts a fleet of Model Context Protocol (MCP) servers shared across your workspace. Install Claude Desktop / Cursor MCP tools once, available to every Maestro chat and every agent in the workspace. Community catalogue surfaces popular MCP servers.
Quick answer
- Workspace-shared MCP server fleet.
- Compatible with Claude Desktop / Cursor MCP tools.
- Available to all Maestro chats and agents in the workspace.
- Community catalogue surfaces popular servers.
Key capabilities
- Hosted MCP server runtime (stdio, HTTP, SSE transports).
- Compatible with the standard MCP tool format from Claude Desktop / Cursor.
- Workspace-shared install — same server, every member, every agent.
- Per-tool scoping for agent profiles.
- Community catalogue of popular servers.
- Custom MCP servers as Room Apps.
How to do it in osFoundry
- Browse the catalogue — Open /community/mcp-servers, pick one (e.g. filesystem, postgres, fetch).
- Install to workspace — One click. The server boots in the workspace fleet; its tools become workspace-shared.
- Scope to agents — In each agent profile, toggle which MCP tools the agent can call.
Use cases
- AI consultant: Pre-install the same set of MCP servers across all client workspaces. Standard tool surface across engagements.
- Internal AI tools team: Publish your company’s internal-API MCP server. Every team member’s agents can use it the moment it’s shared.
Frequently asked questions
Can MCP servers from Claude Desktop be moved here?
Yes — paste the MCP config. The server runs in osFoundry’s fleet and tools become workspace-shared.
What MCP transports are supported?
stdio, HTTP, and SSE.
How is data flow secured?
Tools run in sandboxed runtimes with network ACLs. Per-tool scoping prevents an agent from calling tools outside its allowlist.
Can I write my own MCP server?
Yes — package it as a Room App or follow the MCP spec; install via the catalogue.
Pricing
Per-second compute (the MCP server runs in a container) + per-million tool invocations.
Related features