MCP Tools
By default, LINZA exposes 7 main MCP tools. The main facade for agents is agent_workspace.
The normal user path stays in the default mode: check readiness, index the folder, search, read exact files, inspect counters, and run work through one workflow facade.
In practice, you can ask LINZA in ordinary language: check the folder, show the map, find related material, or prepare a context pack. The names below are a reference for what the MCP client calls during the work.
Main Tools
| Tool | Purpose |
|---|---|
guide_next_steps | Show the next safe step |
agent_workspace | One entry point for map, inbox, review, apply, teach, grow, connect, memory, traces, and context export |
index_all | Update the Markdown-folder index |
search | Search notes through semantic + lexical search |
read_file | Read an exact Markdown file |
get_stats | Show sidecar counters |
scan_vault | Check the folder without writing |
agent_workspace Actions
agent_workspace carries the main workflow inside one tool. It can show the map, handle incoming material, return review cards, apply selected IDs, search accepted memory, and export context.
| Action | Purpose |
|---|---|
doctor | Check sidecar, index, inbox, review gates, and calibr readiness |
map | Read-only working-folder map |
ingest_artifacts | Import incoming material into the sidecar |
analyze_inbox | Analyze incoming artifacts |
review_next | Show next aw-* cards |
apply_review_items | Preview or apply selected aw-* IDs |
history | Show accepted and revoked decisions |
revoke_approval | Softly revoke an accepted decision by approval_id |
teach | Select seed cards |
grow | Select new cards from accepted examples |
connect | Explain a relation between two nodes |
search_memory | Search artifacts and accepted memory |
export_context | Markdown context packet |
record_trace | Store an agent-work trace |
analyze_trace | Trace metrics |
review_calibr | calibr cards for review |
How To Ask LINZA
If the path is unknown, ask LINZA to find material on the topic. Under the hood, this usually starts with search, then opens an exact file through read_file.
If you need the folder state, ask LINZA to check the folder and show the map. This corresponds to agent_workspace(action="doctor") and agent_workspace(action="map").
For any sidecar or compact YAML write, ask for review cards and a preview first. Internally, this is review_next, then apply_review_items with dry_run=true. A real write needs exact IDs and explicit dry_run=false.
For a handoff to a new session, ask LINZA to prepare a context pack. Context packs are saved in .linza/context-packs.
Additional Mode
For development and audit, enable the full tool list:
$env:LINZA_TOOL_SURFACE="advanced"It adds low-level operations: index_file, write_file, patch_properties, search profiles, tag audit, reports, build_review_apply_queue, approve_review_queue_items, show_flow, create_context_pack, and diagnostics.
write_file in this mode writes Markdown only and requires explicit allow_overwrite=true to replace an existing file. The normal LINZA path uses review cards and sidecar records.
If the next step is a dedicated semantic ontology for a knowledge base, open NOUZ.