MCP server
loremfile runs as a Model Context Protocol server, so an agent can find a fixture, read its URL and measured properties, and check a file it already has. It returns URLs and metadata, never file bytes: the agent, or your test, fetches the URL. The server talks over stdio and reads the manifest once per process from loremfile.dev.
Configure your client
With Node, the npm package runs the server:
{ "mcpServers": { "loremfile": { "command": "npx", "args": ["-y", "loremfile", "mcp"] } } }
With Python, uvx runs the PyPI package without an
install:
{ "mcpServers": { "loremfile": { "command": "uvx", "args": ["loremfile", "mcp"] } } }
Both are the same server, tool for tool: the two packages ship one tool definition file. It
speaks protocol 2026-07-28 and 2025-11-25, so clients of either era work.
Tools
All three are read-only.
| Tool | Arguments | Returns |
|---|---|---|
list_fixtures |
format, tag, max_bytes, limit, offset |
One page of published fixtures, each with path, url, format, mime, bytes, tags and description, plus total and next_offset. |
describe_fixture |
path |
One fixture's manifest entry: url, mime, bytes, sha256, size_class, tags, description, props and added_in. |
verify_file |
path, file |
Whether a local file is the published one: ok, changed or missing, with both hashes and sizes. The file is read to hash it and goes nowhere. |
Registry
The server is listed in the
MCP Registry
as io.github.kumarprabhashanand/loremfile, pointing at the npm and PyPI packages above.
Without an agent, the same files are a URL away: see getting started.