The design is about output volume. A raw HAR is unreadable and enormous; this returns one line per request, and only the entries you name come back with headers and bodies. Listing the domains first is the intended move — it shows you what to exclude before the request list is worth reading.
A reader for HAR files on your disk. It flattens a capture into one line per request with a short hash, and those hashes are how you ask for the full detail later.
- Every request as a compact line — hash, status, method, URL — with query parameters shown or hidden
- Filtering by status code, HTTP method, URL pattern, or by excluding domains you do not care about
- Full headers and request and response bodies for specific entries, addressed by hash, one at a time or several at once
- A list of every unique domain in the capture, which is how you find the noise worth filtering out
No account and no key — it reads a .har file from a path you give it. Run it with npx -y @naotaka/mcp-har-server@latest over stdio, or from a build with node build/index.js. The package also installs mcp-har-cli for the command line and mcp-har-sse for an SSE transport.
One command — npx -y @naotaka/mcp-har-server@latest