MCP Interviewer CLI
MCP Interviewer checks an MCP server before an agent connects: provider limits, tool tests and a Markdown report.
- 158 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- With `--test` the tool calls the checked server's tools and runs it as a child process.
- Tests and evaluation send the server's data to the chosen model.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/mcp-interviewerDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- With `--test` the tool calls the checked server's tools and runs it as a child process.
- Tests and evaluation send the server's data to the chosen model.
This is third-party code. Review the repository files before installing.
What it does
`mcp-interviewer` is a Python command-line tool. It starts the given MCP server as a child process, reads its capabilities and checks them against provider limits, for example OpenAI: at most 128 tools per request and names up to 64 characters. With `--test` a language model, using your key, builds a test plan and calls the server's tools while collecting statistics. The `--judge-tools` and `--judge-test` flags add an experimental model evaluation. The output is a Markdown report `mcp-interview.md` and a `mcp-interview.json` file with raw data. Report sections are chosen with `--reports`.
Who it is for. MCP server authors and maintainers who want to know how a server behaves in a model's hands.
Good fit when
- A report on the server's tools is needed before a release.
- A tool set is checked against OpenAI limits.
- Two servers with similar functions are compared by call statistics.
- CI should catch tool schemas that break.
Not a fit when
- Server security is the target rather than model usability.
- The server can change data and there is no container for testing.
- A model verdict would be taken as final without reading it.
Example request
Check my MCP server with mcp-interviewer in a container, without functional tests, and show limit violations and a short report.Limitations
With `--test` the tool itself calls the server's tools, so run the server in a container with no access to real data. `--test` and model evaluation need your own model and key, for example `--model gpt-4.1`. Model evaluations are experimental: verify every conclusion by hand. Basic constraint checking and the report work without a model. The tool runs the checked server's command on your machine, so check only servers you trust. The last commit in the repository dates from September 2026.
How to disable. Remove the package with `pip uninstall mcp-interviewer` and delete generated report files. Revoke the model key if you issued it only for testing.
FAQ
What is checked without a model?
Provider constraints (`--constraints`, for example `OTC`, `ONL`, `ONP`) and the report contents: server info, tools, resources, prompts.
What does the --test flag do?
A model builds a test plan from the server's capabilities and calls its tools. `--model` and a safe environment are required.
Where is the result written?
Into `mcp-interview.md` and `mcp-interview.json`; sections are tuned through `--reports`.
Can it be called from Python?
Yes, the package installs as a dependency through `uv add mcp-interviewer` or `pip install mcp-interviewer`.
Is it safe to run the checked server?
Run it in a Docker container: the tool executes the server command in a child process and, with `--test`, calls its tools.
Discussion
No questions or reviews yet.
Related
CLI toolsThe official MCP server debugger: web UI, automation CLI and terminal UI in one package
The npx @wong2/mcp-cli command runs MCP servers from a Claude Desktop config, lists their tools and calls them from the terminal.
MCP server: an agent sends GET, POST, PUT, DELETE and PATCH to your REST API and gets back status, headers and body
kogiQA MCP gives an agent a browser that finds page elements without selectors, for debugging web apps and writing e2e tests
sunpeak tests any MCP server in local replicas of the ChatGPT and Claude hosts and builds MCP Apps with React
MCP server for a device test bench: an agent leases a board, connects and runs jmp commands on real or virtual hardware.
README in short
The tool interviews an MCP server: it checks constraints, optionally tests tools with a model and writes a report. Microsoft researchers describe the idea in an article on tool-space interference in the MCP era.