dekart 0.3.0

Dekart CLI for auth, MCP tool discovery, calls, and file upload flows.

dekart-cli

Standalone Dekart CLI.

Install

pip install -e .

After install, command is:

dekart --help

Config and auth

dekart config --url http://localhost:3000
dekart init

Config and token are stored under:

• ~/.config/dekart/config.json
• ~/.config/dekart/token.json

After successful dekart init, the CLI can optionally install local snapshot capability. You can control this behavior with:

dekart init --local-snapshot ask      # default
dekart init --local-snapshot install  # install right after auth
dekart init --local-snapshot skip     # skip prompt/install

MCP tools

dekart tools --json
dekart call --name create_report --args '{}'
dekart upload-file --file /tmp/result.csv --file-id <file-id>

You can also stream upload content from stdin:

bq query --use_legacy_sql=false --format=csv 'SELECT 1 AS x' \
  | dekart upload-file --stdin --file-id <file-id> --name result.csv --mime-type text/csv

--stdin is staged to a temporary file and uploaded in parts, so it avoids loading the full stream into memory. Total-size limits are enforced by Dekart server configuration.

Snapshots

Render report PNG snapshots for agent verification:

dekart snapshot --report-id <report-id> --out ./snapshot.png

dekart snapshot uses local headless rendering when local snapshot is enabled. It uses remote snapshot only when local snapshot is disabled (or when you pass --remote-only).

Manage local snapshot capability:

dekart snapshot-local status
dekart snapshot-local install
dekart snapshot-local uninstall

Optional full browser cleanup on uninstall:

dekart snapshot-local uninstall --purge

Local release hook (main branch)

Install repo hooks:

./scripts/install_hooks.sh

On main, pre-push will:

1. Bump minor version in pyproject.toml and commit it.
2. Build and publish to PyPI with twine.
3. Stop the first push intentionally.

Then run push again:

git push origin HEAD
git push origin HEAD

Auth:

• If PYPI_API_TOKEN is set, hook uses token auth.
• Otherwise it uses local Twine auth (~/.pypirc, keyring, or prompt).