English | Русский

TAUSIK MCP — Tool Reference (v1.5)

124 tools for AI agents (117 project + 7 brain; v1.5 actual count, asserted via len(TOOLS) on both servers). The MCP surface covers everything an agent does day-to-day. A few CLI-only commands have no MCP equivalent — they are operator / maintenance verbs that don't belong in an agent loop: skill rebuild, skill bundle, fts optimize, db prune, audit vendors/research, config set/show, push-ok, run, doc extract/constants, hud, suggest-model, hygiene archive --confirm. For the agent's working set, prefer MCP tools over shell calls — they are atomic, return structured data, and keep your context cleaner.

Optional codebase-rag server adds 7 tools (search_code, find_symbol, …). It is enabled separately during bootstrap and is NOT part of the main 124 count — total with it is 131 tools.

Two MCP servers live in this project:

• tausik-project — project-scoped tools (99): tasks, sessions, knowledge, stacks, roles, gates, skills, exploration, audit, doctor, verify, usage logging.
• tausik-brain — cross-project Shared Brain tools (7).

There is also an optional codebase-rag server documented at the bottom.

Verify-First Contract (v1.5)

Heavy quality gates (pytest, tsc, cargo, phpstan, javac, js-test, terraform-validate, helm-lint, kubeval, hadolint, ansible-lint) live on a dedicated verify trigger. The MCP workflow:

tausik_task_start(slug=…)        # QG-0
… work on code …
tausik_verify(task_slug=…)        # heavy: subprocess gates → caches green
tausik_task_done(slug=…, ac_verified=True)   # lightweight: cache lookup

tausik_task_done will refuse to close the task if the verify cache is missing or stale — it returns a structured failure with explicit remediation. Opt-out for CI: set {"task_done": {"auto_verify": true}} in .tausik/config.json so the heavy gates fire inside task_done like in pre-v1.5 releases.

Terminology: Verify / QG glossary distinguishes supported opt-out, QG bypass (not available for task_done), verify-cache bypass, and the pytest test shim.

Status, Health, Metrics

Tool	Description	Required Parameters
tausik_health	Health check: version, DB, tables	—
tausik_self_check	MCP-server freshness: startup time, watched-module mtime snapshot vs current on-disk mtimes, drift_detected flag, stale modules with delta_seconds, sibling MCP project server count. Call from /start to catch silent-hang precursors (gotchas #77/#79/#80).	—
tausik_status	Project overview: tasks, session, epics. Optional compact: true → one-line JSON (default text unchanged).	compact (optional)
tausik_doctor	4-group health (venv + DB + MCP + skills + drift)	—
tausik_metrics	SENAR metrics: Throughput, FPSR, DER, Dead End Rate, Cost/Task	—
tausik_usage_event_log	Append manual row to usage_events (does not update session aggregates)	tokens_input, tokens_output, tokens_total, cost_usd
tausik_search	Full-text search across tasks, memory, decisions	query

Tasks

Tool	Description	Required Parameters
tausik_task_add	Create task (optionally in a story)	slug, title
tausik_task_quick	Quick creation with auto-slug	title
tausik_task_start	Start work (QG-0: requires goal + AC + negative scenario)	slug
tausik_task_done	Complete (QG-2: ac_verified=true, scoped pytest, verify cache). Returns structured JSON: blocking_failures, per-gate results, cache status.	slug
tausik_task_show	Full task information	slug
tausik_task_list	List tasks with filters (status enum: planning,active,blocked,review,done)	—
tausik_task_update	Update fields (title/goal/AC/scope/notes/stack/complexity/role/tier/call_budget)	slug
tausik_task_plan	Set plan steps	slug, steps[]
tausik_task_step	Mark step as completed	slug, step_num
tausik_task_log	Append journal entry	slug, message
tausik_task_logs	Read structured logs (filter by phase)	slug
tausik_reason_step	RENAR reasoning step (intent|premise|action|verification)	slug, kind, content
tausik_task_replay	Chronological task timeline (logs + reasoning + events + verification)	slug
tausik_task_block	Block task	slug
tausik_task_unblock	Unblock	slug
tausik_task_review	Move to review	slug
tausik_task_delete	Delete task	slug
tausik_task_move	Move to another story	slug, new_story_slug
tausik_task_next	Pick next task by score	—
tausik_task_claim	Claim task (multi-agent)	slug, agent_id
tausik_task_unclaim	Release task	slug

tausik_task_done parameters

• ac_verified — required for QG-2
• evidence — inline AC verification log (replaces a separate task_log call)
• no_knowledge — confirm no knowledge to capture (suppresses warning)
• relevant_files[] — files modified; drives scoped pytest gate (basename match → tests/test_<file>.py). Empty list with non-empty original → gate skipped (no false-positive on full suite). Verify cache (10 min TTL) skips re-runs with same files_hash.

There is no --force on task_done — QG-2 cannot be bypassed. task_start does have --force to bypass session capacity, with audit trail.

tausik_task_done structured response

tausik_task_done returns JSON for agent workflows:

• stage flags (plan_complete, ac_verified, gates_passed)
• per-gate results (gates[])
• blocking_failures[] with gate, files, output, and remediation hints
• warnings[], cache_status, and final ok

Pre-1.4 there was a parallel tausik_task_done_v2 alias for the structured-JSON variant. v14b-task-done-rename-drop-v2 consolidated both into the single tausik_task_done returning the structured JSON above — there is no _v2 suffix. The Verify-First Contract is honoured on every path.

Sessions

Tool	Description	Required Parameters
tausik_session_start	Start session	—
tausik_session_end	End session	—
tausik_session_extend	Extend active-time limit beyond 180 min	—
tausik_session_current	Current active session	—
tausik_session_list	List sessions	—
tausik_session_handoff	Save handoff data	handoff (object)
tausik_session_last_handoff	Get handoff from previous session	—
tausik_session_open (v1.5)	Compound RPC: session start + status + handoff + active/blocked tasks + self_check in one envelope. Powers /start Phase 1.	—

Session limit is gap-based active time (paused after 10-min idle gap), not wall clock. See session-active-time.md.

Hierarchy (Epics and Stories)

Tool	Description	Required Parameters
tausik_epic_add	Create epic	slug, title
tausik_epic_list	List epics	—
tausik_epic_done	Complete epic	slug
tausik_epic_delete	Delete (cascade: stories + tasks)	slug
tausik_story_add	Create story in epic	epic_slug, slug, title
tausik_story_list	List stories	—
tausik_story_done	Complete story	slug
tausik_story_delete	Delete (cascade: tasks)	slug
tausik_roadmap	Tree: epic → story → task	—

Knowledge

Tool	Description	Required Parameters
tausik_memory_add	Save to project memory	type, title, content
tausik_memory_search	Full-text search in memory	query
tausik_memory_list	List entries (filter by type)	—
tausik_memory_show	Show entry by ID	id
tausik_memory_delete	Delete entry	id
tausik_memory_block	Compact markdown: recent decisions + conventions + dead ends (for /start re-injection)	—
tausik_memory_compact	Aggregate recent task_logs (phases + top words + top files)	—
tausik_memory_archive (v1.5)	Soft-archive memory rows older than a duration (90d / 12w / 2m / 1y). Dry-run unless confirm: true.	before (string), confirm (bool, optional)
tausik_memory_dedupe (v1.5)	List near-duplicate memory pairs above a similarity threshold (read-only).	threshold (float, optional), limit (int, optional)
tausik_decide	Record an architectural decision	decision
tausik_decisions_list	List decisions	—

Memory types: pattern, gotcha, convention, context, dead_end.

Graph Memory

Tool	Description	Required Parameters
tausik_memory_link	Create edge between nodes	source_type, source_id, target_type, target_id, relation
tausik_memory_unlink	Soft-invalidate edge (never deletes)	edge_id
tausik_memory_related	Find related nodes (1–3 hops)	node_type, node_id
tausik_memory_graph	List edges with filters	—

Relation types: supersedes, caused_by, relates_to, contradicts.

Dead Ends and Explorations

Tool	Description	Required Parameters
tausik_dead_end	Document a failed approach	approach, reason
tausik_explore_start	Start time-boxed investigation	title
tausik_explore_end	End investigation	—
tausik_explore_current	Current investigation	—

Quality Gates and Verification

Tool	Description	Required Parameters
tausik_gates_status	Status of all gates (by stack)	—
tausik_gates_enable	Enable gate	name
tausik_gates_disable	Disable gate	name
tausik_verify	v1.5 Verify-First: run heavy gates (pytest, tsc, …) and cache green in verification_runs. After that tausik_task_done reads the cache and closes instantly.	task_slug

Available gates: pytest, ruff, mypy, bandit, tsc, eslint, go-vet, golangci-lint, cargo-check, clippy, phpstan, phpcs, javac, ktlint, filesize, tdd_order. Stack-scoped gates auto-enable based on detected stack; universal gates (filesize, tdd_order) apply to all stacks.

tdd_order is disabled by default. Enable with tausik_gates_enable name=tdd_order.

Stacks

Tool	Description	Required Parameters
tausik_stack_list	List built-in + custom stacks	—
tausik_stack_show	Resolved stack: gates per language + override info	stack
tausik_stack_export	Export resolved declaration as JSON	stack
tausik_stack_diff	Diff between built-in and user override	stack
tausik_stack_reset	Remove user override at .tausik/stacks/<stack>/	stack
tausik_stack_lint	Validate user-override stack.json files	—
tausik_stack_scaffold	Create .tausik/stacks/<name>/{stack.json,guide.md} skeleton	name

DEFAULT_STACKS: 25 entries (python, fastapi, django, flask, react, next, vue, nuxt, svelte, typescript, javascript, go, rust, java, kotlin, swift, flutter, laravel, php, blade, ansible, terraform, helm, kubernetes, docker). Custom stacks via .tausik/config.json → custom_stacks.

Roles

Tool	Description	Required Parameters
tausik_role_list	List roles	—
tausik_role_show	Show role profile	slug
tausik_role_create	Create role (optionally extends a base profile)	slug, title
tausik_role_update	Update role metadata	slug
tausik_role_delete	Delete role	slug
tausik_role_seed	Bootstrap rows from harness/roles/*.md + existing task usage	—

Role storage is hybrid: SQLite metadata + harness/roles/{role}.md profile markdown. Roles on tasks remain free-text.

Periodic Audit (SENAR Rule 9.5)

Tool	Description	Required Parameters
tausik_audit_check	Check whether audit is overdue	—
tausik_audit_mark	Mark audit as completed	—

Skills

Tool	Description	Required Parameters
tausik_skill_list	List skills: active, vendored, available	—
tausik_skill_install	Install skill from repo (clone + copy + deps)	name
tausik_skill_uninstall	Uninstall skill completely	name
tausik_skill_activate	Activate installed skill	name
tausik_skill_deactivate	Deactivate skill (keep files)	name
tausik_skill_repo_add	Add TAUSIK-compatible skill repo (third-party URLs need force)	url, optional force
tausik_skill_repo_remove	Remove skill repo	name
tausik_skill_repo_list	List repos and available skills	—
tausik_skill_catalog	Discovery: list skills offered by configured/cloned repos (name, category, description)	optional repo, optional as_json

Cross-Project Queue (CQ)

Tool	Description	Required Parameters
tausik_cq_publish	Publish a cross-project event	payload
tausik_cq_query	Query cross-project queue	—

Multi-agent and Maintenance

Tool	Description	Required Parameters
tausik_team	Tasks grouped by agent	—
tausik_events	Audit log (events)	—
tausik_update_claudemd	Update dynamic section in CLAUDE.md	—
tausik_fts_optimize	Optimize FTS5 indexes	—

Shared Brain (tausik-brain, 7 tools)

Tool	Description	Required Parameters
brain_search	Search the Notion-backed brain (FTS over local mirror)	query
brain_get	Get a brain record by id	id, category
brain_store_decision	Store a cross-project decision	name, decision
brain_store_pattern	Store a cross-project pattern	name, description
brain_store_gotcha	Store a cross-project gotcha	name, description
brain_draft_artifact	Dry-run artifact publish (taxonomy + scrub + classifier risk; no Notion write)	kind
brain_cache_web	Cache a web result for token reuse	name, url, content

The tausik-brain MCP server runs config-agnostic at startup and reads registry from .tausik-brain/ configuration. The total tool count for this server is 7 (verified via len(TOOLS) in harness/claude/mcp/brain/tools.py).

Brain config requirements

When brain.enabled=true in .tausik/config.json, ALL of the following must be set or tausik_decide (and other brain-routing operations) will return a ⚠ ... saved LOCALLY ONLY — brain mirror BLOCKED warning and skip the Notion mirror:

• brain.database_ids.decisions, database_ids.patterns, database_ids.gotchas, database_ids.web_cache — all four Notion database UUIDs.
• brain.notion_integration_token_env — env var name (default NOTION_TAUSIK_TOKEN) that must resolve to a non-empty token via env, .tausik/.env, or brain.notion_integration_token in config.

tausik doctor surfaces validation errors as a Brain config warning row. The fastest fix is tausik brain init (interactive wizard) or set brain.enabled=false to opt out cleanly. After fixing the config, run tausik brain move --to-brain to migrate decisions/gotchas/patterns that were saved locally during the misconfiguration window.

Codebase RAG (separate optional MCP server)

Tool	Description	Required Parameters
search_code	Search project code via RAG index	query
search_knowledge	Search project knowledge base	query
reindex	Reindex the codebase	mode (incremental/full), max_seconds (soft limit, full only). v1.5: stderr progress every 100 files; truncated=true on timeout.
rag_status	RAG index status	—
archive_done	Archive completed tasks	—
cache_web_result	Cache web search result for reuse	query, content
search_web_cache	Search cached web results	query

These are not part of the main 105 count — they belong to the optional codebase-rag server.

Launching the Tausik MCP Server

The bootstrap step generates IDE-specific MCP launchers under harness/<ide>/mcp/. Claude Code reads .claude/settings.json (auto-generated). To regenerate IDE assets and MCP wiring, run python bootstrap/bootstrap.py from your TAUSIK checkout (or python .tausik-lib/bootstrap/bootstrap.py when using the submodule layout). Use python bootstrap/bootstrap.py --refresh only to rewrite .tausik/config.json (e.g. after setting TAUSIK_MODEL_PROFILE) without copying skills/scripts — it does not regenerate .mcp.json files.