Recommended Free Tools
A live MCP endpoint is not automatically discoverable, easy for clients to inspect, or visible to operators. First identify what “nobody could see” means: whether the server owner cannot list its tools, colleagues cannot find or configure it, or operators cannot see whether it is working. Each is a different problem with a different check.
What does “visible” mean for an MCP server?
MCP server visibility has at least three layers. A server may be reachable at its configured endpoint, yet absent from a directory; it may be listed in a directory, yet fail when a client connects; or clients may use it successfully while operators lack useful health and usage data.
As an Amazon Associate I earn from qualifying purchases.
| Visibility question | What to check | What the check does not prove |
|---|---|---|
| Can a client reach the server and enumerate capabilities? | Connect to the actual endpoint with an appropriate MCP client or MCP Inspector; check initialization and tools/list. |
That other users can discover it, or that production calls are monitored. |
| Can intended users find and configure it? | Check the public registry or your organization’s server inventory, as appropriate, and verify the server’s descriptive and distribution metadata. | That the listed endpoint is currently healthy. |
| Can the operator tell whether it is working? | Check service telemetry and operational dashboards for signals such as traffic, errors, latency, and resource use. | That a directory listing or one successful test reflects ongoing health. |
The MCP Registry announcement describes the public registry as a catalog and API for publicly available servers, intended to help clients discover them. It also notes compatible sub-registries for differentiated needs. The announcement, dated September 8, 2025, described the registry as available in preview at that time; check its current publishing status and requirements before relying on a specific submission workflow.
Free tools Windows power users keep installed
One-click scans. No signup required.
If you cannot see the server’s tools, test the endpoint
When the owner cannot enumerate capabilities, begin with the server configuration and connection path—not a public directory. Use an MCP client or Inspector configured for the actual server endpoint and credentials. The Inspector documentation describes inspecting protocol methods including initialize and tools/list; its configuration guide explains how to configure a server connection. See the Inspector README and MCP server configuration guide.
#1 Best Overall
- Confirm the target. Check that the client or Inspector is pointed at the endpoint you intend to diagnose, using the transport and connection settings required by that server.
- Use the required credentials. Supply the authentication configuration that the endpoint expects. A test with missing or different credentials does not establish that the intended client path works.
- Check initialization. Determine whether the connection reaches the MCP initialization exchange. If it does not, investigate endpoint reachability, transport configuration, and authentication before treating this as a discovery problem.
- Request the tool list. Check whether
tools/listsucceeds and returns the expected tools. If initialization works but listing fails or returns no expected tools, focus on the server’s capability/listing path and configuration.
Inspector is a developer diagnosis tool for a server you can already address. Connecting to it does not publish it to users or provide persistent production monitoring.
If coworkers cannot find or configure it, check discovery
If you can connect and enumerate the tools but your intended users cannot locate the server, check where that audience is expected to discover it. Public availability and company-only availability call for different distribution choices.
For a publicly available server
The official MCP Registry announcement describes a public catalog and API for discovery. Registry metadata, including the server.json format, describes a server for publishing and discovery; consult the server.json format specification for the format’s details. Keep descriptions and distribution details accurate and current, and verify the registry’s present publishing process rather than assuming preview-era instructions still apply. A listing is not a health check.
For a company-only server
Do not assume that public listing is the right route for an internal service. Organizational inventories can help members register and find servers within a particular environment. Google Cloud documents registering and managing MCP servers and tools through Agent Registry; Microsoft documents registering and discovering servers in Azure API Center. These are examples of platform-specific internal discovery, not interchangeable or universally available directories. Choose based on your organization’s platform, access controls, authentication requirements, and who owns keeping the inventory entry current.
Rank #3
If users connect but operators cannot tell whether it works, add observability
A successful connection test is a point-in-time check; a registry entry is descriptive metadata. Neither answers whether calls continue to succeed, how long they take, or how heavily the service is used. For operations, instrument the running service or use an observability service that can expose the signals relevant to your deployment.
Google Cloud’s Agent Registry documentation describes an MCP server view with latency, traffic, errors, token spend, and system utilization. That is a vendor-specific feature example, not a default capability of every MCP server or hosting platform. See Google Cloud’s MCP server and tools management documentation for its stated scope.
Rank #4
- Server 2022 Standard 16 Core
- Traffic: whether calls are reaching the service and when usage occurs.
- Errors: whether requests fail, so operators can distinguish “unused” from “used but failing.”
- Latency: how long requests take, rather than merely whether the endpoint responds.
- Resource use: whether the service’s compute or other resources are under pressure.
- Token spend, where relevant: a cost or usage signal supported by some platform views, not necessarily available in your deployment.
Use the symptom to choose the next check
- No tools visible to the owner: test the configured endpoint and credentials with an MCP client or Inspector; inspect initialization and
tools/list. - Tools visible to the owner, but not to coworkers: verify the intended public registry or internal inventory, plus its server metadata and user configuration path.
- Users can connect, but operations are unclear: check service telemetry or a platform view that provides persistent health and usage signals.
- A directory shows the server, but clients fail: treat the listing and runtime connection as separate checks; verify the actual endpoint, transport, and authentication.
The title alone does not establish which of these failures occurred, what transport or host was used, or whether the endpoint was healthy. The key is to test the layer that matches the missing kind of visibility instead of treating “live,” “listed,” “inspectable,” and “monitored” as synonyms.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
