ember
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| EMBER_HOST | No | Model server address (host). | 127.0.0.1 |
| EMBER_PORT | No | Model server address (port). | 8765 |
| EMBER_MODEL | No | Model to use: flash (9B) or full (27B). | flash |
| EMBER_DEVICE | No | Device to run the model on: auto, mps, or cpu. | auto |
| EMBER_AUTOSTART | No | Let the MCP server start the model server on demand. | 1 |
| EMBER_MODEL_DIR | No | Run weights from this directory instead of the pinned cache. | |
| EMBER_STATE_DIR | No | Where the pid file and logs live. | Application Support |
| EMBER_MAX_LENGTH | No | Token cap per request; 0 derives it from the model (the model's maximum is 262144). | 0 |
| EMBER_SERVER_URL | No | Where the MCP server sends requests. | http://127.0.0.1:8765 |
| EMBER_START_TIMEOUT | No | Seconds to wait for the model server to start. | 300 |
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": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| adviseA | Get ember's read on a situation, as calibrated probabilities. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| ember-guide | Full playbook for the advise tool: schemas, thresholds, and recipes. |
TDQS
Scored across 1 tool
Only one tool exists, so there is no risk of confusion with other tools; its purpose is clearly stated as getting calibrated probabilistic advice.
The single tool name 'advise' is a clear verb, and no inconsistent conventions exist within the set.
One tool is slightly below the typical 3-15 range, but it is well-scoped for a focused advisory endpoint that handles multiple query modes internally.
The tool covers choice, score, and noul response types, plus image/video evidence, providing a complete surface for its stated advisory purpose.