Read runs

Web UI

Projects, a project's runs (filter by status), and a run page with charts, samples (images, videos, text tables), logs, config, system usage, alerts and artifacts. Pick runs to compare them side by side. Everything updates live. It is made for the phone first: add it to the home screen to install it, and tap the bell to get push notifications on that device.

CLI

runyard runs ls [--project P] [--status S]
runyard runs show RUN            # RUN: id, id prefix or run name
runyard logs RUN [-n 100] [-f]
runyard samples RUN [--out DIR]  # downloads the latest samples
runyard artifacts RUN [--get NAME] [--out DIR]
runyard sync                     # sends runs saved while the server was unreachable
runyard watch DIR [--project P] [--once] [--interval 10]

Every read command takes --json for scripts and agents. With RUNYARD_AUTH on, set RUNYARD_API_KEY for reads too.

MCP for agents

The server is an MCP server at /mcp:

claude mcp add --transport http runyard https://my-server/mcp --header "Authorization: Bearer $RUNYARD_API_KEY"

Tool	What it returns
list_projects, list_runs	projects with run counts; runs with status, config and summary
get_run	status, timing, host, git, config, summary, and per metric its count, last, min and max
get_metrics	curves summarized as a short [step, value] series per key
get_samples	the sample images, as images the agent can look at, and text samples
tail_logs	the last lines the run printed (stderr lines start with !)
compare_runs	runs side by side: the config keys that differ, last/min of shared metrics
list_alerts, list_artifacts	alerts raised by the run or by checks; checkpoints and models
add_note, add_tags	write a verdict or mark a run; need the API key

SKILL.md tells an agent how to use them.

REST API

Reads are GET under /api/v1: projects, runs?project=&status=&tag=, runs/{id} (also an id prefix or name), runs/{id}/metrics?keys=&points=, runs/{id}/logs?after=&tail, runs/{id}/media?key=&kind=&step=, runs/{id}/artifacts, artifacts/{id}/file, alerts?run=&project=, and events (server-sent events, one line per change). Writes need Authorization: Bearer <key>: POST runs, PATCH runs/{id}, and POST of runs/{id}/metrics, logs, media, heartbeat, alerts and artifacts.