Server has 9 tools with complete input schemas and descriptions in French. Naming follows verb_noun pattern (hydra_read_job, hydra_write_job, etc.). Descriptions are moderate length (100-250 chars) and explain WHAT the tool does and WHEN to use it. However, parameter descriptions lack actionable constraints (no enums for file types, no format specs for job_dir paths). Output schemas are not documented, LLMs cannot predict return structure. Error handling is implicit (validation mentioned but no recovery guidance). Security considerations present (atomic writes, validation) but not explicitly surfaced in descriptions. No pagination support documented for list tools despite potentially large result sets.

• Document output schemas for all tools. Example: hydra_list_jobs should return {jobs: [{path: string, valid: boolean, errors?: string[]}], total: number}. This enables LLMs to chain tools and extract needed fields.
• Add pagination to list tools: hydra_list_jobs(limit=20, offset=0) → {jobs: [...], total: 100, has_more: true}. Cap default limit at 20 and document in description.
• Add format constraints to job_dir parameter: 'Relative path (alphanumeric, dash, underscore, slash; no ../ or absolute paths; max 255 chars). Example: my_job or projects/acme/jobs/etl_daily'.
• Add error recovery guidance to descriptions. hydra_delete_job: 'Irreversible. Consider calling hydra_validate_job first to confirm job exists. No undo available.' hydra_run_job: 'If execution fails, check logs in returned journal. Retryable errors: timeout, connection loss. Non-retryable: invalid schema, missing source.'
• Translate all descriptions to English. Replace French descriptions with English equivalents to maximize LLM comprehension and agent compatibility across multilingual deployments.