Contentstack MCP
Connect Claude, Cursor, ChatGPT or any other MCP client to Contentstack with one hosted URL. Choose a profile, sign in, and your agent gets the tools for that job.
Coming from the @contentstack/mcp package?
Profile Hub replaces the stdio package. It covers the same products from one hosted URL and gives each connection a curated toolset instead of full tool groups. See how the two compare.
Connect Claude, Cursor, ChatGPT or any other MCP client to Contentstack with one hosted URL. Choose a profile, sign in, and your agent gets the tools for that job.

Profile Hub runs at a single address. Add it to your client as a remote MCP server.
https://mcp-profile-hub.contentstack.com/api/mcpYou do not install anything or paste an API key. Your client lists the profile's tools once sign-in completes. Steps for each client are in the docs, or use the form on the right to build a ready-made install command.
Optionally create a pre-configured MCP URL here. This way you can connect your stack, brandkit, or launch projects directly in the URL.
Leave a field blank to keep the stack default. Blank values are left out of the URL.

Build context-aware MCPs, not API wrappers
Discover why simply wrapping APIs isn't enough for Model Context Protocol (MCP) servers. Learn how adding project and user context can transform MCPs into dynamic, efficient engines that enhance AI reasoning, reduce errors, and scale across complex environments.

MCP fragmentation, context efficiency, and the rise of curated skills
This article explores the fragmentation of the Model Context Protocol (MCP) in the AI ecosystem, highlighting challenges with context window efficiency, the complexity developers face, and emerging solutions like curated tool subsets and Claude Skills. It offers insights into protocol-layer issues, strategies for improving reliability, and the pragmatic necessity for adaptable system architectures amid ongoing standards wars.
AI will not live in one place, but trust has to
This article discusses the tension between rapidly expanding enterprise AI capabilities and the need for strong governance, introducing a two-layer architecture: off-platform AI for experimental reach and on-platform AI for trusted, governed execution. It explains how combining both layers allows organizations to benefit from flexible, distributed AI while maintaining oversight, brand integrity, and risk management.
We don't test our MCP profiles. We test the agent.
How we test MCP Profile Hub for AI agents: a deterministic tool matrix plus behavior evals for grounding, safety, and correct tool use, gated on every change.
Across the Contentstack APIs and products, an agent connected through a profile can:
What it can actually do depends on the profile. A read-only profile cannot publish.
When no system profile fits, build your own in the MCP Profile Hub Marketplace app. Pick the tools a job needs from the catalog, connect the stack, and copy the install snippet. That is three steps: tool selection, configuration, installation.
The connector URL is safe to share. Identity comes from each person's own sign-in, so a shared profile never gives anyone extra access.
Your Automate workflows and Agent OS agents with an HTTP trigger show up as tools too. One call can run a whole workflow.
Profile Hub replaces the local @contentstack/mcp package. Both cover the same products. They differ in how you install them and how many tools your agent ends up with:
@contentstack/mcp package | MCP Profile Hub | |
|---|---|---|
Installation | Runs locally via npx, configured per machine with credentials or an OAuth session | Hosted at mcp-profile-hub.contentstack.com. Nothing to install, add a URL to your client |
Transport | stdio, so it only works in clients that can run a local process | Streamable HTTP, so it also works in browser-based clients like claude.ai and ChatGPT |
Tool selection | Tool groups (cma, launch, and so on). Each group loads its full toolset, up to 77 tools for CMA alone | Curated per profile. You expose exactly the tools a job needs, nothing more |
Profile Hub signs you in with standard OAuth 2.1 and PKCE. There are no API keys to paste, share or rotate. Every call runs with your own Contentstack roles, so a profile can narrow what you can do but never widen it.
Every tool call is logged. The Executions view in the Marketplace app shows the status, tool, timing and any error for each call.
Frequently asked questions
No. The public URL works on its own. You need the app to build custom profiles, save configuration or review executions.
No. You sign in through the browser with OAuth.
No. profile_id and org_uid are identifiers. A stack API key on the URL does not grant access without your sign-in.
No. Each person connects with their own login and roles.
The package runs locally over stdio and loads full tool groups. Profile Hub is hosted, uses Streamable HTTP and exposes only the tools a profile needs.
Change the URL parameters, or remove the server from your client and add it again to choose new ones at sign-in.
https://mcp-profile-hub.contentstack.com/api/mcp