docs: explain how MCP servers actually launch (stdio vs SSE, Inspector) - #4091
docs: explain how MCP servers actually launch (stdio vs SSE, Inspector)#4091TanishkaMarrott wants to merge 1 commit into
Conversation
Closes a recurring point of confusion (modelcontextprotocol#521): developers run `uvx mcp-server-git` directly, see no output, and think the server is broken. Added a new 'Understanding How Servers Are Launched' subsection that explains: - stdio servers wait on stdin — appearing to hang is expected - the MCP client spawns the server; you don't run it yourself - stdio vs SSE/HTTP transport comparison table - MCP Inspector as the recommended local testing tool Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
/i-promise-this-is-not-a-new-server |
|
@cliffhall — this is a docs-only fix to README.md (21 lines added, no new servers). It closes #521 which has been open for a while. The CI failure is a permissions error on the bot's side, not an issue with the PR itself — the branch only touches README.md. Would appreciate a quick look when you have bandwidth! |
|
+1, this would have saved me time. I was just writing READMEs for some |
|
Thanks @MukundaKatta — good call. I've added an explicit note right above the standalone |
Summary
Closes #521 — addresses the recurring confusion around running MCP servers.
npx/uvxexamples and the client config sectionuvxcommandnpx @modelcontextprotocol/inspector uvx mcp-server-git) as the recommended way to test a server locally without wiring up a full clientNo existing content changed. The new subsection slots in cleanly between the two existing subsections.
Test plan
🤖 Generated with Claude Code