The whole point of a config diagnostic is showing it to someone else, and MCP configuration is exactly where API keys live. Redacting by default rather than as an option means the report can go into an issue or a chat without a review step. The write path is separated the way it should be — scan, explain, plan, then apply behind an explicit confirmation with backups already written — which matters more here than usual, because the thing being modified is the configuration the agent itself runs on. Context weight as a first-class diagnostic is a good instinct: an over-attached MCP setup degrades every turn quietly, and nothing else measures it.
A diagnostic tool for the MCP configuration itself: it scans local config files, explains what it found, generates a reversible patch plan, and applies it only after explicit confirmation.
- dr_mcp_scan is the one-step check for an agent session — context weight and stale entries among them — and dr_mcp_cleanup returns the highest-value cleanup candidates without applying anything.
- scan_mcp_setup reads local configuration files and returns a redacted diagnostic report, so the output can be shared without leaking credentials.
- explain_issue expands a single diagnostic rather than leaving a finding as a label.
- generate_patch_plan produces a reversible plan and apply_patch_plan applies it only after explicit confirmation, creating backups before it writes.
- export_report writes the redacted report as Markdown, HTML or JSON.
Nothing to supply — it reads the MCP configuration already on the machine.
One command — npx -y @inferensys/dr-mcp server