| Component | Path | Role |
|---|
| Entry | tools/mcp/index.mjs | Bootstrap and stdio transport |
| Server | tools/mcp/server.mjs | McpServer class, JSON-RPC over stdio |
| Infrastructure | tools/mcp/infra.mjs | Platform utils (paths, file I/O) |
| Domain modules | tools/mcp/domain/ | Tool implementations by domain |
All memory operations route through a single unified Orchestrator_memory MCP tool with action-based routing:
| Action | Description | Key Parameters |
|---|
recall | Search memory with scope/type filters | query, layer (all/semantic/episodic/procedural/knowledge), k, section |
learn | Store fact with type classification | key, value, type, source (user_explicit/observed/inferred/external_unverified) |
episode | Record event or reflect on task | event, outcome, op (reflect/blindspot/suggest/patterns) |
procedure | Save reusable workflow | name, steps[], context |
manage | Consolidate, delete, update, seed, index | op (consolidate/delete/update/contradictions/seed/index) |
session | Session state management | op, notes, persist |
system | System info and diagnostics | op |
When layer: "knowledge" is specified, recall routes to passage search over indexed articles:
Orchestrator_memory(action:"recall", query:"<topic>", layer:"knowledge", section:"<optional>")
Records task outcome with trust calibration:
Orchestrator_memory(action:"episode", op:"reflect", agent:"<name>", outcome:"success|partial|failure", worked:"<what>", failed:"<what>")
| memory_contradictions | Detect same-type facts that are similar but conflicting (flags, no auto-merge) | threshold (0-1, default 0.85) |
| memory_delete | Delete memories by pattern (substring match). Destructive. | pattern, dry_run |
| Tool | Description | Key Parameters |
|---|
reflect | Record task reflection; outcome feeds success-weighted learning | task, worked, failed, pattern, outcome (success/partial/failure) |
reflect_suggest | Recall best-scoring past approach for a situation | situation |
blindspot | Record or check blindspots. action=log records, action=check matches context. | action (log/check), area, observation, context |
| Tool | Description | Key Parameters |
|---|
dashboard | Full overview of memory, metrics, patterns, procedures, micro-tools, rules | section (all/memory/metrics/patterns/procedures/tools/rules) |
| Tool | Description | Key Parameters |
|---|
user_profile | Get/update profile, or discover identity from all sources | update, action (get/discover), persist, name |
session_log | Track work duration, rest reminders | action (start/end/check), context |
goal | Set/check/done exit conditions | action (set/check/done/status/clear), condition, max_turns |
loop | Recurring reminders at intervals | action (start/stop/list/clear/check), prompt, interval |
shutdown_ritual | Wind-down helper | action (configure/check/trigger/status/snooze), bedtime |
| Tool | Description | Key Parameters |
|---|
play_music | YouTube/local player with taste learning, playlist, radio mode | action (play/stop/pause/next/prev/like/dislike/radio/...), query, source |
Knowledge operations are accessed through the same Orchestrator_memory tool:
| Operation | Invocation | Description |
|---|
| Index/seed | action:"manage", op:"seed" | Scan knowledge/ markdown, store summaries + chunk bodies |
| Section list | action:"recall", layer:"knowledge", list_sections:true | List available knowledge sections |
| Passage search | action:"recall", query:"...", layer:"knowledge", section:"..." | Semantic passage search over article bodies |
| Tool | Description | Key Parameters |
|---|
zara_self_audit | Validate config integrity: agents vs prompt files, plugin modules, MCP domains | map |
| Tool | Description | Key Parameters |
|---|
user_profile | Get/update user profile, or discover identity from all sources (env, git, OS, memory) | action (get/discover), update, persist, name |
| Tool | Description | Key Parameters |
|---|
project_learn | Extract project knowledge (stack, structure, tests, CI) from workspace and store as scoped facts | path |
| Tool | Description | Key Parameters |
|---|
zara_self_improve | Run self-improvement cycle: observe (gather signals) or orient (prioritize findings) | phase (observe/orient/full) |
| Domain | Tools |
|---|
| Memory | 7 |
| Reflection | 6 |
| Metrics | 3 |
| Session | 5 |
| Music | 1 |
| Knowledge | 4 |
| Audit | 2 |
| Project | 1 |
| Identity | 1 |
| Improve | 1 |
| Total | 31 |
| Removed | Absorbed By |
|---|
memory_stats | dashboard(section: "memory") |
metrics_today | dashboard(section: "metrics") |
micro_tools | dashboard(section: "tools") |
workflow_rules | dashboard(section: "rules") |
patterns | dashboard(section: "patterns") |
zara_evolve_status | dashboard(section: "all") |
blindspot_log | blindspot(action: "log") |
blindspot_check | blindspot(action: "check") |
user_identity | user_profile(action: "discover") |
chm2md | CLI-only: node tools/chm2md.mjs |
chm2md_improve | CLI-only: node tools/chm2md.mjs --improve |