MCP integrations, tool by tool
Every integration guide here answers the same three questions in the same order: does a server actually exist for this tool, what can an agent do once it is connected, and what will bite you in production. Vendor quick-starts stop after step two.
The pattern that repeats across every platform: the setup is rarely the hard part. The hard parts are permission scope (an integration account with the narrowest role you can get away with), result-set size (a query that returns 50k rows will blow through your context window long before it blows through the API quota), and the gap between what the API exposes and what the MCP server actually wraps.
Start read-only. Every guide below tells you how to do that for its specific platform, because 'read-only' means something different in ServiceNow than it does in Postgres.
If you only change one thing after reading any guide here: give the agent its own credentials with its own role, not yours.
42 articles · page 3 of 3
MCP Server for Kubernetes: the options
Five real MCP servers that talk to Kubernetes, matched to what you actually run — plus the read-only config that keeps an agent off your prod cluster.
MCP Server for Documentation: The Real Options
Five ways to stop your agent guessing at APIs — and which one to actually install.
FAQ
Is there an MCP server for every SaaS tool?
No. Coverage is uneven: developer tools and databases are well served, while many enterprise platforms only have community servers or none at all. Each guide here states plainly what exists for that specific tool, including when the honest answer is 'use the REST API directly'.
Do I need admin rights to connect a tool over MCP?
Usually no, and you should not use them. Most integrations work with a scoped API token or an OAuth app limited to the objects you need. Create a dedicated integration account rather than reusing your own credentials, so the agent's blast radius is separate from yours.
Why does my agent time out on large queries?
Because tool results go into the model's context window. A query returning tens of thousands of rows will exhaust context or hit a transport timeout long before it hits an API limit. Every integration guide here gives the practical row/result caps for that platform.
Other topics
Looking for a server rather than an article? Best MCP servers by task