The rlg-mcp server exposes three tools (tail_log, filter_log, summarize_errors) but the source code provided contains NO tool definitions, schemas, parameter descriptions, or implementation details visible. The Dockerfile mentions the tools exist, but no actual MCP tool registration, JSON schemas, or descriptions can be verified in the provided source. Cargo.toml and Makefile show workspace structure but not the rlg-mcp crate implementation. Without visible tool definitions, this evaluation must apply the hard scoring rule: tools cannot score above 50 per-tool, and per-tool scores with no visible schemas or descriptions must be 0 for schema and 0 for description.

• Provide the actual rlg-mcp crate source code (crates/rlg-mcp/src/lib.rs or main.rs) showing tool registration with MCP framework.
• Define a JSON Schema for each tool's input parameters. Example: tail_log should specify log_file (string), lines (number with min/max), timeout_seconds (number). All parameters must have type and description fields.
• Add substantive descriptions (50-200 chars) for each tool. E.g., tail_log: 'Stream the last N lines from a log file in real-time, useful for monitoring application events or debugging failures. Returns lines as an array of strings with timestamps. Does not modify the file.'
• Add descriptions to every input parameter. E.g., 'log_file: Path to the log file to tail. Required. Must be readable by the server process.' and 'lines: Number of lines to retrieve. Integer, 1-1000. Default 20.'
• Document output schema including field names, types, and structure. E.g., tail_log returns {lines: string[], total_lines: number, last_offset: number, is_truncated: boolean}.
• Implement error handling with recovery guidance. If a file is not found, return 'File not found: /var/log/app.log. Readable log files in /var/log: app.log, app-error.log' to enable the agent to self-correct.
• Add enums for constrained inputs. E.g., if filter_log accepts a 'level' parameter, define it as enum with values: 'trace', 'debug', 'info', 'warn', 'error', 'fatal'.
• Consider pagination support in tail_log and filter_log if large log files are common. Add offset/page_token and limit parameters with documented minimums (1) and maximums (e.g., 10000 lines).