Integrations: run, import, mcp

vouch run

Record a run of any command — the entry point for non-Python experiments.

$ vouch run ID [--dep PATH]... [--input PATH]... [--out PATH]... [--values FILE] \
      [--prefix P] [--row-key C] [--stats] -- CMD...

Flag	Meaning
ID	the run id
--dep PATH	code the run depends on (file or directory); repeatable
--input PATH	data the run reads; repeatable
--out PATH	a file the run writes (an artifact); repeatable
--values FILE	read values from this results file instead of $VOUCH_VALUES
--prefix	prefix for every key
--row-key	for tabular values: the column naming each row
--stats	lists of numbers become Stats
-- CMD...	everything after -- is the command to run

$ vouch run cifar_vit_jl --dep src/ --dep configs/vit.yaml --input data/cifar10.npz \
      --out results/vit.csv -- julia train.jl --model vit

Full details: Non-Python experiments.

vouch import

Register an existing results file as a run, with honest, reduced provenance.

$ vouch import FILE --run ID [--prefix P] [--producer PATH]... [--command TEXT] \
      [--row-key C] [--stats] [--root DIR]

Flag	Meaning
FILE	the results file
--run ID	the run id (required)
--prefix	prefix for every key
--producer PATH	code that produced the file (file or dir); repeatable
--command TEXT	the command that produced it (declared, not observed)
--row-key	for tabular files: the column naming each row
--stats	lists of numbers become Stats

$ vouch import results/imagenet_eval.json --run imagenet_eval --prefix imagenet \
      --producer experiments/eval_imagenet.py --producer src/models/ \
      --command "python experiments/eval_imagenet.py --split val"

vouch mcp

Serve search/cite/compare/check/… to any MCP client over stdio.

$ vouch mcp

Read-only except that compare may write a definition when asked. Built on newline-delimited JSON-RPC 2.0 with the standard library — no extra dependencies. vouch init --agents mcp registers it in .mcp.json.

Full tool list and protocol details: MCP server.