MCP Servers
Which MCP server creates the most value for your most frequent task — and is that the one you've actually configured?
MCP servers give agents structured access to external tools and data — databases, APIs, file systems, code. The protocol is the bridge; choosing the right servers is the discipline. This hub is the value view: which servers to adopt and why. To build a server, see MCP Server Engineering.
Where to Start
- Choosing tools? Go to the MCP Tool Radar — team-specific load lists, a 6-gate evaluation checklist, and a governance protocol for keeping your toolset lean.
- Browsing what exists? The Server Catalog — a reference list of available servers by category.
- Want to understand how MCP works, or build one? The engineering view — MCP Server Engineering — covers architecture, primitives, transports, and token economics.
Why Selection Is the Job
Every tool you load costs tokens before the agent asks a single question. Loading everything is prohibitively expensive; loading nothing is crippling. The radar and profiles exist to find the balance — maximum capability at the lowest token cost your workflow can carry.
Context
- pairs-with MCP Server Engineering — the build view of the same protocol
- instance-of AI Toolkit — MCP servers are one aisle of the toolkit
- depends-on Agent Protocols — where MCP fits the broader protocol stack
- up AI Toolkit — the toolkit hub
Questions
Which MCP server category creates the most value for your most frequent task — and is that category what you've actually configured?
- At what server count does context-window management become the binding constraint?
- Which server would you drop first if your session budget were cut in half?