Skip to content

MCP Tools

Tools are the main interface of NOUZ. Through them an AI agent does not merely read a folder of files; it works with a knowledge base as a structure: nodes, links, levels, domains, bridge candidates, and drift signals.

The philosophy is simple: every action should be explicit. One tool reads a note, another shows its graph position, another suggests metadata, and another recalculates the composition of the base. Automation proposes; the human, or an agent following your rules, decides.

NOUZ provides up to 15 tools through the Model Context Protocol. Availability depends on the mode:

#ToolMeaningLUCAPRIZMASLOI
1read_fileread a note with YAML and refresh it in the index
2write_filewrite a note with cycle checks and synchronized links
3list_filesinspect the file map without loading full text
4get_childrentraverse down the graph: what this node contains
5get_parentstraverse up the graph: where this note belongs
6format_entity_compactget a compact formula for a node's position
7index_allbuild the local index of files, YAML, and links
8suggest_metadatapropose domain, level, tags, bridges, and warnings
9suggest_parentsfind possible parents by semantic similarity
10embedtest an embedding for arbitrary text
11calibrate_coresbuild reference vectors for semantic domains
12recalc_signsrecalculate automatic classification for the base
13recalc_core_mixaggregate domain composition bottom-up and reveal drift
14process_orphansprocess files without metadata and propose fills
15add_entitycreate a new entity with initial metadata

Reading and Writing

read_file

Read a Markdown file and return YAML frontmatter together with content. The YAML fields include type, level, sign, artifact_sign, parents, and tags. Reading also re-indexes the file in the database, so later suggestions use fresh state.

write_file

Create or update a note with YAML frontmatter. Checks the graph for cycles before writing and syncs simple parent links parents with detailed link metadata parents_meta. This is the final-action tool: read and suggest first, then write.

list_files

List indexed files with filters by level, sign, subfolder, or no_metadata.


Graph Navigation

get_children

Get descendants of a node from the DAG index: direct and transitive.

get_parents

Get parent links for a file. Returns entity and link_type for each connection.

format_entity_compact

Compact formula for a node's position: (children)[node]{parents}. The formula displays link types that help keep the structure readable: hierarchy, semantic, and temporary.


Semantics PRIZMA / SLOI

suggest_metadata

Analyzes a file and suggests sign, artifact_sign, level, tags, bridges, and hierarchy errors. Suggestions return proposed: true: the decision stays with you.

suggest_parents

Finds parent candidates for orphan notes by cosine similarity. Candidates below parent_link_threshold are discarded; candidates from the same domain win ties.

embed

Generate an embedding for arbitrary text. Useful for debugging similarity and manual checks.


Database Maintenance PRIZMA / SLOI

index_all

Re-index all markdown files. With with_embeddings=true, also recomputes embeddings.

calibrate_cores

Recompute reference vectors from config.yaml. Returns raw and mean-centered pairwise cosine values to check domain separation.

recalc_signs

Reclassify files against current etalons. Updates domain sign sign, automatic sign sign_auto, sign source sign_source, and material type artifact_sign in the database. YAML files are not changed. Use dry_run=true to preview.

recalc_core_mix

Recalculate core_mix bottom-up: L4 gets a profile from text classification, while L3 and L2 aggregate child nodes. Updates the database only.


Automation PRIZMA / SLOI

process_orphans

Finds files with missing signs or missing parents and suggests sign/artifact_sign, tags, and possible parents. Can run as dry_run or write YAML.

add_entity

Creates an entity in one step: writes the file, determines sign/artifact_sign, tags, and with auto_parents=true adds the best parent above threshold.

{Semiotronika}
Telegram · Volnaya Sreda · Email