gel-mcp provides 6 tools for querying Gel databases and accessing examples/rules. Tool names follow verb_noun conventions (list_*, fetch_*, execute_*, try_*). Descriptions are present and reasonably clear (ranging 58-238 chars). Input schemas are properly defined with types and descriptions for all parameters. However, output schemas lack documentation, return types are inferred from code but not declared to the LLM. Error handling is minimal (basic ValueError raises, no recovery guidance). Parameter descriptions are adequate but lack constraint details (e.g., what constitutes a valid slug or rule_name?). Missing are: per-parameter type hints in descriptions, actionable error messages, and explicit output field documentation. Composition is sound, tools are single-purpose and chainable. The server targets domain experts (Gel developers) rather than general users, which justifies some simplifications. The execute_query and try_query tools are particularly well-designed with clear reversibility semantics.
Execute a query and return the result as JSON
Fetch a code or workflow example by its slug
Fetch a rule by its name. E.g. gel.md or gel-python.md
List all available code and workflow examples and their slugs
List all available rules. Rules are Markdown files that contain examples and instructions for AI agents on how to use Gel. They are crucial for correct code generation.
Execute a query in a transaction that gets rolled back, allowing you to test queries without making permanent changes
Output schemas not documented. Return types are inferred from code (list[str], list[Any], str) but LLM receives no schema, forcing it to guess what fields are available and waste tokens parsing unstructured responses.
Parameter descriptions lack actionable constraints. 'slug' and 'rule_name' parameters have descriptions but no hints about valid formats, length limits, or where to find valid values. LLM must infer from names alone.
Error messages are minimal and non-actionable. FileNotFoundError and ValueError are raised with brief text (e.g., 'Rule {rule_name} not found'). No recovery guidance (e.g., 'Call list_rules() to see available rules').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 60 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 42 | - | v1 |
No pagination or result limiting for list_examples and list_rules. If workflows or rules grow large, responses could exhaust context. Best practice: add limit and offset/cursor parameters.
execute_query and try_query accept optional 'arguments' and 'globals' as untyped dicts. No guidance on what keys/values are valid or what they control. LLM cannot validate input before passing.