OLACHEAP · MCP KNOWLEDGE BASE
MCP guides — discover, connect, verify
A practical field guide to Model Context Protocol: what MCP servers do, how to choose and connect one, how to verify a provider, and how to troubleshoot safely.
1. Understand the components
An MCP server exposes tools and resources to a compatible AI client. An agent skill, by contrast, is an instruction or workflow; WebMCP works in a browser context. A directory listing is only a discovery record.
- Identify the AI client and the action you want to automate.
- Check whether the listing is remote (published endpoint) or local (manual installation).
- Read the maintained documentation, transport, required account, permissions and version.
2. Choose the right server
A good fit is determined by capabilities, maintenance and access scope, not an invented popularity score.
- Search by task and read the server details and upstream documentation.
- Compare official source URL, version, remote endpoint and the actual permissions requested.
- Test in a non-production account first. Never assume a directory listing means security certification.
3. Connect remote or local MCP
Remote and local installations follow different trust and setup paths.
- Remote: copy only the published HTTPS endpoint into a client that supports that transport.
- Follow the provider's official authentication flow, such as OAuth; approve the smallest scopes.
- Local: follow the provider's package and runtime documentation, inspect source and dependencies before installation.
4. Verify an MCP provider
Website verification proves control of a particular host at the time of checking. It does not certify the server code or all links on a shared source platform.
- Create a claim for the matching server name and your own website.
- Publish the exact requested DNS TXT record or the exact-host HTTPS proof file.
- Run verification, then review published details. Shared hosts such as github.com cannot be claimed by owning a single repository.
5. Troubleshoot a broken connection
Separate discovery errors from authentication, network and tool permission failures.
- Check the server endpoint and supported transport without posting credentials.
- Re-authorize the account when a token has expired; check scopes only after confirming ownership.
- Check upstream status and client logs. Keep secrets, access tokens and private prompts out of public issue reports.