Mcp Config And Status — Claude Code Guide

Configuration describes a connection, not a secret

MCP configuration should record server, transport, scope, auth boundary, and available operations. Real tokens and passwords must not enter the repository, diff, logs, or teaching examples.

Credentials should come from the environment or a secret manager, while the configuration remains reviewable.

Project-scoped configuration location

workflow-kit/
└── .claude/
    └── mcp/
        └── issue-tracker.json

Transport can be local or remote

Transport	How it works	Typical case
stdio	Claude launches a local process through standard streams	Script or local utility
http	Claude calls an already running server over the network	Team tracker, docs, or monitoring

Schematic transport examples

{"name":"docs-local","transport":"stdio","command":"python","args":["scripts/docs_server.py"]}

{"name":"issue-tracker","transport":"http","endpoint":"${ISSUE_TRACKER_URL}"}

Configuration scope and auth scope differ

user — a server for several personal projects;
local — an uncommitted experiment;
project — a shared repository workflow;
plugin — capability delivered with a package.

Concept	Question answered
Config scope	Who sees settings and where they are stored?
Auth scope	Which operations can the token or account perform?

Status check and recovery

Inspect the configured server and transport.
Confirm that credentials are resolved outside source files.
List available tools without invoking writes.
Check the smallest read-only request.
Record errors, scope, and fallback.
Disable or remove the server when its boundary is unclear.

Canonical lesson source · JavaRush

Open JavaRush material