MCP GUIDES
Create, host and connect MCP servers.
Start with the task you need to finish. Each answer leads to a walkthrough or reference with the next steps.
What is an MCP server?
An MCP server exposes tools and data through the Model Context Protocol. A compatible AI client discovers the tools and their input schemas, then calls them when needed. The tools can query your database or call an existing API.
How do I create an MCP server?
On Glxymesh, start a project with the gxmesh CLI and add a tool. Describe a single HTTP request in YAML, or write a run function in Go, TypeScript or Python. For AI-assisted creation, describe the tool in the console, review its draft and tests, then approve its access before publishing.
How do I host an MCP server?
Managed hosting gives your project an HTTPS endpoint and handles the runtime and authentication. With Glxymesh, gxmesh push builds, tests and deploys the tools. Self-hosting gives you control of the infrastructure and responsibility for its deployment, access controls and monitoring.
How do I connect an AI client to MCP?
Copy the project endpoint from the console's Connect page, add it using your client's remote MCP setup, and complete the OAuth authorization. Check that the client supports Streamable HTTP with OAuth, then enable the tools you want it to call. Exact settings vary by client.
How do I secure MCP tools and API keys?
Grant each tool only the hosts and credentials it needs. Glxymesh stores keys as project secrets and adds HTTP credentials at the egress gateway, so the tool code and calling AI client do not need the key. Review generated tools, input validation and write actions before publishing.
Why is my MCP server not connecting?
First check the exact endpoint, supported transport and completed OAuth authorization. If connected tools are missing, confirm they are deployed and enabled. If a tool appears but its call fails, inspect the invocation outcome, allowed hosts and secret bindings instead of repeating client setup.
Ready for a working endpoint? Follow the first deployment tutorial.