clef-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| CLEF_MODEL | No | Model name sent with every request. | clef |
| CLEF_API_KEY | No | Bearer token. Required when the host is ollama.com or *.ollama.com; optional otherwise. The API key is never logged or included in error messages or tool output. | |
| CLEF_BASE_URL | No | Ollama server base URL. | https://ollama.com |
| CLEF_TIMEOUT_MS | No | Request timeout in milliseconds (positive integer). | 60000 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
clef-mcp exposes one tool, decide, backed by Ollama's /v1/systemone decision endpoint. Use it when you need a fast, probability-calibrated judgement about a piece of text, JSON data, or images: pick one label (choice), estimate how likely a yes/no statement is true (noul), or grade on an ordered scale (score). Batch every question about the same state into one call. The model is fixed by the server configuration and cannot be chosen per call.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| decideA | Ask an Ollama decision model to answer one or more structured questions about a state, in a single call. Answers are probabilities, never free text, so you can threshold them. Inputs:
Every question has a type, instructions (text, or a JSON object or array, explaining what to decide) and criteria:
Use it for classification, routing, triage, moderation and rubric grading. Do not use it to generate text. |
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 1 tool
There is only one tool, so there is nothing to confuse it with. Within the tool, the three question types (choice, noul, score) are clearly delineated with distinct criteria and answer shapes, so an agent can select the right mode without ambiguity.
A single, clear verb name ('decide') that accurately reflects the operation. There are no competing conventions to clash, and the naming reads naturally for the domain.
The server exposes exactly one tool, which is thin by the usual 3-15 guidance, though it is a genuinely single-capability wrapper around a decision model and the batch 'questions' parameter absorbs what would otherwise be many tools. It is borderline rather than mismatched, since no second capability obviously belongs here.
The tool covers the core decision surface well: batch questions, three answer modes, optional images, and probability outputs for thresholding. Minor gaps remain, such as model discovery/selection, health or capability introspection, and any guidance on limits or failure modes, which an agent would have to work around.