open source · npm published · 2026
FeathersJSMCPServer
an mcp server for the feathersjs framework. it gives AI coding assistants four tools over stdio — semantic search over a bundled knowledge base, and github pr integration for docs. no http port, no database, nothing to run in the background.
setup
one command. guided wizard.
run the init command and a cli wizard handles editor selection and config path. no manual json editing required.
$ npx feathersjs-mcp-server@latest init
┌ feathersjs mcp server · setup wizard
│
│ ? which editor would you like to configure?
│ › Claude Code
│
│ ? config install path?
│ › ~/.claude/ (default)
│
└ configuration complete
✔ wrote ~/.claude/claude_code_config.json
✔ feathersjs-mcp-server registered
restart your editor to activate the server.
$
what it does
four tools, one knowledge base
all four tools use the same bundled feathersjs index. search works offline. the three pr tools need github — they only run when you pass them a token.
search_docssearches the bundled feathersjs knowledge base fully offline. the index ships with the package, so there's no network call.
submit_documentationopens a github pr to add a new doc entry to the knowledge base. needs a github token and network access.
update_documentationopens a github pr to update an existing entry when docs go stale or the api changes.
remove_documentationopens a github pr to remove an entry from the index. keeps things clean without manual file editing.
demo
see it in action
compatibility
works where you already code
all over stdio — no daemon to manage, no port to forward.
my role
project manager
pmi came in as pm. that meant scoping the work, writing the task breakdown, running the weekly syncs, and staying the main contact between the team and the client.
the job was to scope it, keep it moving, and get it to a publishable state. we broke the work into pieces, ran weekly syncs, and made sure nobody was waiting on anything.
i wasn't just writing the code... i was making sure the people who were had what they needed.
