Right now, the inspector has a CLI feature as well as a series of tests.
However, they're not currently run in CI. For the moment, they are only
triggered on file changes in the CLI folder, as AFAICT it is currently a
separate code base.
This commit removes the prompt argument from the completion state.
The argument is deleted from the context after the completion is handled.
This prevents unexpected behavior and ensures a cleaner state.
This commit adds a `context` argument to the completion handlers in `PromptsTab`, `ResourcesTab`, `useCompletionState`, and `useConnection`.
The context argument allows passing additional information to the completion engine, which can be used to improve the quality of the completions. The context is passed as a dictionary of key-value pairs.
The `ResourcesTab` component was also updated to pass `templateValues` as the `context` to the completion handler.
The `useConnection` hook was updated to include the context in the completion request sent to the server.
Attempts to execute this command will fail: npx @modelcontextprotocol/inspector --cli http://127.0.0.1:8000 --transport http. The error indicates that --method is a mandatory parameter: "Method is required. Use --method to specify the method to invoke. Failed with exit code: 1". That's why the right command may look like this: npx @modelcontextprotocol/inspector --cli http://127.0.0.1:8000 --transport http --method tools/list.
At the moment, the CLI tool expects that remote MCP servers, either
using SSE or streamable HTTP transport, have certain endpoints or
fragments -- `/sse` and `/mcp`, respectively. This is not part of the
MCP spec, nor should it be expected.
I also added a check to make sure a URL is provided if STDIO isn't
picked, erroring out if not.
Integrate JSON validation with copy JSON functionality:
- Maintain forwardRef structure for validation exposure
- Combine copy button with validation method
- Both features working together without conflicts
