CLI reference¶
Generated from the command-line help at build time.
mcparser¶
Usage: mcparser [OPTIONS] COMMAND [ARGS]...
LLM agents for dependency parsing with CoNLL-U files.
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --version Show the version and exit. │
│ --install-completion Install completion for the current shell. │
│ --show-completion Show completion for the current shell, to copy │
│ it or customize the installation. │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Commands ───────────────────────────────────────────────────────────────────╮
│ validate Validate a CoNLL-U file. Exits with status 1 if there are errors. │
│ run Score and refine a prediction file with the agents, over several │
│ rounds. │
│ score Only run the scorer: give every changed or unscored cell a │
│ confidence. │
│ init Create a prediction file from raw text or a gold CoNLL-U file, │
│ then apply the frequency baseline. │
│ evaluate Compare predictions with gold annotation (the CoNLL 2018 metrics: │
│ precision, recall and F1). │
│ settings Print the settings (defaults, or CONFIG with defaults filled in) │
│ as YAML. │
│ mcp The CoNLL-U MCP server. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser init¶
Usage: mcparser init [OPTIONS] {INPUT}
Create a prediction file from raw text or a gold CoNLL-U file, then apply the
frequency baseline.
╭─ Arguments ──────────────────────────────────────────────────────────────────╮
│ * INPUT <path> Text (one sentence per line), YAML with 'text' │
│ fields, or CoNLL-U (annotation is removed, │
│ segmentation kept). │
│ [required] │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ * --output -o <path> CoNLL-U file to │
│ create. │
│ [required] │
│ --source -s <str> Trusted example │
│ sources, │
│ [NAME=]PATH[:ro[… │
│ Repeatable. │
│ --baseline --no-baseline Fill in what the │
│ trusted sources │
│ make obvious (no │
│ LLM). │
│ [default: │
│ baseline] │
│ --force Overwrite OUTPUT │
│ and its │
│ review/history │
│ files. │
│ --max-sentences <int range> Only use the │
│ [x>=1] first N sentences │
│ (after any │
│ --sentence │
│ filter). │
│ --skip <int range> Skip the first N │
│ [x>=0] sentences of │
│ INPUT (e.g. to │
│ start a new │
│ batch). │
│ [default: 0] │
│ --help Show this message │
│ and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser run¶
Usage: mcparser run [OPTIONS] {prediction}
Score and refine a prediction file with the agents, over several rounds.
╭─ Arguments ──────────────────────────────────────────────────────────────────╮
│ * prediction <path> Prediction CoNLL-U file (edited in place). │
│ [required] │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --source -s <str> Trusted example │
│ sources, │
│ [NAME=]PATH[:ro[… │
│ Repeatable. │
│ --corpus <str> [NAME=]PATH[,PAT… │
│ unannotated text │
│ (one sentence or │
│ paragraph per │
│ line) that agents │
│ can search for │
│ how a word is │
│ used. Repeatable. │
│ --config -c <path> Settings YAML │
│ (see `mcparser │
│ settings`). │
│ --model <str> One model for │
│ every role │
│ (overrides │
│ --strong-model │
│ and │
│ --efficient-mode… │
│ --strong-model <str> Model for the │
│ demanding roles │
│ (default │
│ anthropic:claude… │
│ --efficient-mod… <str> Cheaper model for │
│ the │
│ --efficient-roles │
│ (default │
│ anthropic:claude… │
│ --light-model <str> Cheapest model, │
│ for the │
│ --light-roles │
│ (default │
│ anthropic:claude… │
│ --efficient-rol… <str> Comma-separated │
│ roles that use │
│ the efficient │
│ model (default │
│ upos,feats,score… │
│ --light-roles <str> Comma-separated │
│ roles that use │
│ the light model │
│ (default xpos). │
│ --strong-thinki… <str> Reasoning for │
│ strong roles: │
│ off, on, minimal, │
│ low, medium, │
│ high, xhigh │
│ (default: the │
│ model's own). │
│ --efficient-thi… <str> Reasoning for │
│ efficient roles │
│ (same values as │
│ --strong-thinkin… │
│ --light-thinking <str> Reasoning for │
│ light roles (same │
│ values as │
│ --strong-thinkin… │
│ --batch-sentenc… <int range> Maximum sentences │
│ [x>=1] per agent task │
│ (default 8). │
│ --model-for <str> ROLE=MODEL │
│ override, e.g. │
│ lemma=anthropic:… │
│ --roles <str> Comma-separated │
│ roles to run: │
│ segmentation, │
│ upos, xpos, │
│ lemma, feats, │
│ syntax, head, │
│ deprel, deps. │
│ --rounds <int range> Maximum rounds. │
│ [x>=1] │
│ --joint-syntax --separate-synt… One agent sets │
│ HEAD and DEPREL │
│ together │
│ (default), or │
│ separate head and │
│ deprel agents. │
│ --final-score --no-final-score Re-score the │
│ cells changed in │
│ the last round │
│ (default off). │
│ --max-cost <float range> Stop starting │
│ [x>=0.01] agent tasks once │
│ the estimated │
│ cost reaches this │
│ many dollars (run │
│ again to │
│ continue). │
│ --corpus-tool --no-corpus-tool Let agents call │
│ corpus_find │
│ (default) or give │
│ them only the │
│ corpus evidence │
│ in the prompt. │
│ --threshold <float range> Refine cells │
│ [0<=x<=1] below this │
│ confidence. │
│ --concurrency <int range> Sentences │
│ [x>=1] processed at │
│ once. │
│ --sentence <str> Only work on this │
│ sent_id. │
│ Repeatable. │
│ --max-sentences <int range> Only use the │
│ [x>=1] first N sentences │
│ (after any │
│ --sentence │
│ filter). │
│ --report <path> Write the run │
│ report as JSON. │
│ --help Show this message │
│ and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser score¶
Usage: mcparser score [OPTIONS] {prediction}
Only run the scorer: give every changed or unscored cell a confidence.
╭─ Arguments ──────────────────────────────────────────────────────────────────╮
│ * prediction <path> Prediction CoNLL-U file (edited in place). │
│ [required] │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --source -s <str> Trusted example sources, │
│ [NAME=]PATH[:ro[:CONFIDEN… │
│ Repeatable. │
│ --corpus <str> [NAME=]PATH[,PATH...]: │
│ unannotated text (one │
│ sentence or paragraph per │
│ line) that agents can │
│ search for how a word is │
│ used. Repeatable. │
│ --config -c <path> Settings YAML (see │
│ `mcparser settings`). │
│ --model <str> One model for every role │
│ (overrides --strong-model │
│ and --efficient-model). │
│ --efficient-model <str> Cheaper model for the │
│ --efficient-roles (default │
│ anthropic:claude-sonnet-5… │
│ --efficient-thinking <str> Reasoning for efficient │
│ roles (same values as │
│ --strong-thinking). │
│ --batch-sentences <int range> [x>=1] Maximum sentences per │
│ agent task (default 8). │
│ --model-for <str> ROLE=MODEL override, e.g. │
│ lemma=anthropic:claude-so… │
│ --concurrency <int range> [x>=1] Tasks run at once. │
│ --sentence <str> Only work on this sent_id. │
│ Repeatable. │
│ --max-sentences <int range> [x>=1] Only use the first N │
│ sentences (after any │
│ --sentence filter). │
│ --report <path> Write the run report as │
│ JSON. │
│ --help Show this message and │
│ exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser evaluate¶
Usage: mcparser evaluate [OPTIONS] {prediction} {gold}
Compare predictions with gold annotation (the CoNLL 2018 metrics: precision,
recall and F1).
╭─ Arguments ──────────────────────────────────────────────────────────────────╮
│ * prediction <path> Predicted CoNLL-U. [required] │
│ * gold <path> Gold CoNLL-U. [required] │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --json Print JSON instead of a table. │
│ --predicted-only Ignore gold sentences that are not in the │
│ predictions. │
│ --annotated-only Only score predicted sentences that have │
│ been parsed (every HEAD set). │
│ --calibration Also show accuracy by recorded confidence │
│ (Conf in MISC). │
│ --strict Also score all features and relation │
│ subtypes (the official metrics ignore them). │
│ --official <path> Also run the official conll18_ud_eval.py at │
│ this path (sentences it would refuse are │
│ repaired first). │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser validate¶
Usage: mcparser validate [OPTIONS] {path}
Validate a CoNLL-U file. Exits with status 1 if there are errors.
╭─ Arguments ──────────────────────────────────────────────────────────────────╮
│ * path <path> CoNLL-U file to validate. [required] │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --source -s <str> Trusted sources whose tagsets flag unseen values. │
│ --limit <int> Maximum issues to print. [default: 100] │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser settings¶
Usage: mcparser settings [OPTIONS]
Print the settings (defaults, or CONFIG with defaults filled in) as YAML.
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --config -c <path> Settings YAML (see `mcparser settings`). │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser mcp serve¶
Usage: mcparser mcp serve [OPTIONS]
Serve CoNLL-U sources over MCP (stdio for Claude Desktop, or streamable HTTP).
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ * --source -s <str> [NAME=]PATH[:ro|rw[:CONFID… │
│ Repeat for several sources. │
│ Default mode is ro. │
│ [required] │
│ --transport <stdio|streamable-http> MCP transport. │
│ [default: stdio] │
│ --host <str> Host for streamable-http. │
│ [default: 127.0.0.1] │
│ --port <int> Port for streamable-http. │
│ [default: 8000] │
│ --read-only Register only the read-only │
│ tools. │
│ --root-key <str> MISC attribute holding the │
│ word's root. │
│ [default: Root] │
│ --name <str> Server name shown to MCP │
│ clients. │
│ [default: conllu] │
│ --alignment -a <str> NAME=SOURCE_A,SOURCE_B,PAT… │
│ word alignments between two │
│ sources. Repeatable. │
│ --corpus <str> [NAME=]PATH[,PATH...]: │
│ unannotated text (one │
│ sentence or paragraph per │
│ line) that agents can │
│ search for how a word is │
│ used. Repeatable. │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
mcparser mcp config¶
Usage: mcparser mcp config [OPTIONS]
Print the JSON to add to Claude Desktop's claude_desktop_config.json.
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ * --source -s <str> [NAME=]PATH[:ro|rw[:CONFIDENCE]]. Repeat for │
│ several sources. Default mode is ro. │
│ [required] │
│ --read-only Register only the read-only tools. │
│ --root-key <str> MISC attribute holding the word's root. │
│ [default: Root] │
│ --name <str> Server name shown to MCP clients. │
│ [default: conllu] │
│ --alignment -a <str> NAME=SOURCE_A,SOURCE_B,PATH[:ro|rw[:CONFIDEN… │
│ word alignments between two sources. │
│ Repeatable. │
│ --corpus <str> [NAME=]PATH[,PATH...]: unannotated text (one │
│ sentence or paragraph per line) that agents │
│ can search for how a word is used. │
│ Repeatable. │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯