once mcp

Expose Once's graph and memory queries to a coding agent

Synopsis#

plaintext
once mcp [OPTIONS]

Description#

Speaks the Model Context Protocol over standard input and output so a coding harness can discover schemas and starters, validate and edit typed graphs, inspect or execute annotated scripts, run graph capabilities, and query project evidence without scraping prose. Mounts inspection tools by default; pass --allow-run to expose manifest editing, test, build, run, and runtime session tools.

Options#

FlagValueDefaultDescription
--workspace<DIR>Workspace root the agent tools resolve targets against. Defaults to the value of the global -C/--directory flag (or the current directory)
--allow-run(flag)falseAdvertise and allow state-changing editing and execution tools
-C, --directory<DIR>Project root. Defaults to the current directory; the cache lives under <project>/.once/. Mirrors make -C
--format<FORMAT>humanOutput format for Once's structured data (cache stats, run/exec trailers). Defaults to a human-readable rendering; pass json or toon to get machine-parseable output for scripting and for agent consumers
-v, --verbose(flag)0Increase log verbosity. Repeat for more (-v: info, -vv: debug, -vvv: trace). Overridden by RUST_LOG
-q, --quiet(flag)falseSuppress human-mode success and progress trailers. Errors and the structured envelope of --format json/toon still print. Mirrors the -q flag of common build tools
--list(flag)falsePrint the command surface at the current command depth
--memory-limit<SIZE>Maximum memory scheduling budget for local actions. Defaults to two thirds of the memory visible to the host or container