Loading ToolBench
MCP server for integrating Claude with local Ollama models
Three tools with complete input schemas and reasonable descriptions. Tool names follow verb_noun pattern (ollama_generate_text, ollama_chat, ollama_embed_text). Descriptions are 150-200 chars, exceeding the 10-char minimum but lacking specificity on when to use vs. alternatives. All parameters have types and descriptions. However, descriptions are generic guidance rather than LLM-optimized prompts. No output schemas documented. Error handling is minimal, tools throw generic 'Ollama X failed' messages without recovery guidance. Missing parameter constraints (e.g., temperature range 0.0-2.0 stated but not enforced; max_tokens unbounded). No idempotency or retry guidance.
Have a conversation with local Ollama for SIMPLE Q&A, factual questions, or basic explanations that don't require deep reasoning. Prefer for routine queries to save Claude tokens. AVOID for complex analysis, nuanced discussions, or tasks requiring sophisticated reasoning.
Generate text embeddings using local embedding models like nomic-embed-text. Ideal for batch embedding tasks, semantic search, similarity comparisons, and clustering. Use this for routine embedding generation to save Claude tokens.
Generate text using local Ollama for SIMPLE, token-efficient tasks like basic content, error messages, placeholder text, boilerplate code, or routine documentation. Use instead of Claude for non-analytical text generation. AVOID for complex reasoning, analysis, or creative writing that requires nuanced understanding.
No output schemas documented. LLMs cannot infer what fields to expect from responses (e.g., does ollama_generate_text return {text, tokens_used, model}?). Forces LLMs to guess and parse unstructured output.
Error messages are generic ('Ollama generation failed: <error>') with no recovery guidance. LLMs cannot determine if the error is retryable, user-fixable, or fatal. No actionable next steps provided.
Parameter constraints stated in descriptions but not enforced in schema. Temperature range '0.0-2.0' is text-only; max_tokens has no minimum/maximum bounds. LLMs may pass invalid values (negative tokens, temperature > 2.0).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Descriptions lack specificity on tool selection criteria. 'Use instead of Claude for non-analytical text generation' is vague. When should an LLM choose ollama_generate_text vs ollama_chat? No clear decision boundary.
No idempotency guarantees or retry guidance. If an LLM retries ollama_generate_text after a timeout, will it produce duplicate outputs? No documentation of side effects or safe retry behavior.