MCPBundles
Overview
MCPBundles is a hosted MCP platform: streamable HTTP, workspace-scoped credentials, and a large catalog of integrations you enable in the dashboard. The Hub endpoint (https://mcp.mcpbundles.com/hub/mcp) exposes platform tools—catalog and credential management, bundle enablement, agents, and code execution—so your client isn’t limited to one tool call at a time when a workflow needs composition, aggregation, or branching.
Key features
- Hub MCP URL — one remote connection for platform operations; authenticate with OAuth where your client supports it, or a personal API key (
mb_…) via headers on the Hub request. - Large integration catalog — CRM, billing, data stores, dev tools, and more; you choose what to enable per workspace.
- Programmatic multi-tool runs — the Hub code execution tool runs sandboxed Python that can call multiple enabled MCP tools in a single run (sequential or concurrent orchestration, filtering, and aggregation in-process) instead of forcing the model to round-trip every intermediate result. Suited to multi-step workflows and large intermediate payloads you don’t want back in the chat context.
- Agents — trigger and manage heartbeat runs for longer, multi-step agent workflows against your workspace tools (with cancellation and execution history where your plan supports it).
- Workspace boundaries — credentials and enablement are scoped to the workspace; members share the same Hub and enabled bundles.
CLI
Install the open-source mcpbundles CLI from PyPI (pip install mcpbundles). It connects to the same Hub with stored or env-based credentials, supports tool discovery and invocation from the terminal (mcpbundles tools, mcpbundles call …), embedded code execution (mcpbundles exec …), and an interactive shell for scripted or automation workflows—useful when you want programmatic access outside a chat UI. Connection and usage details are covered in the docs and repo: GitHub — thinkchainai/mcpbundles.
Docs
Product and setup guides (including Cursor) live under Documentation.
FAQ
Is the Hub local or remote? Remote — HTTPS streamable HTTP; no Docker or npx required for the hosted endpoint.
What belongs on this listing? The Hub MCP URL only. Per-bundle MCP URLs are separate listings.
How does “multi-tool composed” differ from normal MCP? The model or an automation can use Hub code execution to run one Python program that invokes many tools and returns a compact result, instead of serial one-off tool calls for every step.
Server Config
{
"mcpServers": {
"mcpbundles": {
"url": "https://mcp.mcpbundles.com/hub/mcp",
"headers": {
"Authorization": "Bearer YOUR_MCPBUNDLES_API_KEY"
}
}
}
}