Repository navigation
Help us decide what mcp-mock should be #1
LuisPerisVidal
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
👋 Welcome! Help us decide what mcp-mock should be
Hi everyone,
We are the team behind llm-mock, a mock server for LLM provider APIs: you change the
baseURL, keep your code, and test without a real API key.We want to build the same thing for the Model Context Protocol: a mock MCP server you can point your client, host or agent at, with deterministic replies and no real servers or credentials behind it.
This time we are starting with the community instead of the code. The repository is empty on purpose. Before we write anything, we want to know what you need.
What we have in mind
bun start.MCP is not a REST API, so we cannot copy llm-mock as it is. Sessions, capability negotiation, server-initiated requests and stdio all need their own answers, and we would rather find them with you.
Tell us about your case
You do not need to answer everything. Even one line helps.
npx/bunx, or a free hosted instance.Design questions we have not settled
We have opinions, but no decisions yet:
_metafields in the request, a fixture file loaded at startup, or a control endpoint. Which would fit your tests?echo, or also servers that copy the tool catalogs of popular ones such as GitHub, Slack or filesystem?How to take part
We will read everything, summarize the conclusions in this thread, and publish the scope of the first version before we start building.
Thanks for helping us build the right thing 🙌
All reactions