KIHACHI Brain
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| LYRIA_MODEL | No | The Lyria model to use for audio generation, e.g. lyria-3.5. | |
| GEMINI_API_KEY | No | Google AI API key required for the generate_audio tool. Not passed to tool inputs or JSON outputs. | |
| KIHACHI_MEMORY_PATH | No | Path to the memory file to persist memory across restarts. If unset, memory is held only in-process. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| logging | {} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| helloA | Return a simple health-check greeting. |
| generate_songspecB | Generate a SongSpec via Brain, filling omitted values from knowledge. |
| create_project_from_songspecC | Build a ProjectPlan, optionally including its arrangement. |
| create_ableton_planC | Translate a ProjectPlan JSON value into an Ableton plan. |
| execute_live_requestC | Attempt an approved Live request through the configured adapter boundary. |
| prepare_ableton_handoffA | Validate an Ableton plan before a future Live handoff. |
| request_live_executionA | Prepare one approval-gated Live mutation request without executing it. |
| generate_audioC | Generate one audio target through the configured Google Lyria boundary. |
| review_songspecC | Review a SongSpec through the Brain review boundary. |
| remember_songC | Remember a SongSpec and optional ReviewResult. |
| search_memoryC | Search memory by genre and optional review quality filters. |
| orchestrate_songC | Run generation, review, memory, and project preparation. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 12 tools
Most tools map cleanly to distinct pipeline stages: spec generation, review, audio generation, project planning, Ableton planning, and memory operations. The pairs execute_live_request/request_live_execution and create_ableton_plan/prepare_ableton_handoff create mild ambiguity, but their descriptions clarify intent. Overall, an agent can usually select the right tool without much trouble.
All tools except hello follow a clear snake_case verb_noun pattern using domain-specific objects like songspec, project, plan, audio, and memory. hello is a minor exception, reading more like a interjection than a verb_noun action, but it is semantically a health check. This is very close to fully consistent naming.
Twelve tools is well within the expected range for a focused domain and maps to a coherent music-generation lifecycle. Each tool serves a meaningful role whether that is orchestration, individual pipeline stages, or memory access. There is no obvious bloat or redundancy.
The surface covers the main workflow: generate, review, audio generation, project planning, Ableton planning, approval-gated execution, memory storage/retrieval, and orchestration. Gaps like direct retrieval/update/delete of remembered artifacts or editing existing plans would require workarounds but are not dead ends. Core user journeys are coherent and actionable.