trajirctl exposes 6 tools with complete input schemas and descriptions, but suffers from critical naming and composition issues. Tool names lack action verbs (status, export, import, verify, nodes list/show), they read as nouns or subcommands rather than imperative actions. Descriptions are adequate (60 - 120 chars) but generic; they state WHAT without WHEN or WHY. Parameters are well-typed with descriptions, but the tool set violates single-responsibility: 'nodes' combines list and show as subcommands rather than separate tools. No output schemas are documented. Error handling is absent from descriptions. The server is a CLI wrapper, not an MCP-native design, tool definitions appear inferred from CLI dispatch logic rather than explicitly registered.

• Rename tools with action verbs: 'get_trajectory_status', 'export_trajectory', 'import_package', 'verify_package', 'list_nodes', 'show_node'. This makes intent explicit and improves LLM tool selection.
• Split 'nodes' into two independent tools: 'list_nodes' (returns array of node summaries with id, kind, sequence, timestamp) and 'show_node' (returns detailed node info). Document output schemas for both.
• Enhance descriptions with WHEN/WHY: e.g., 'Get trajectory status to check node count and seal status before export. Returns JSON with node_count, seal_count, and counts by kind.' Include prerequisites (e.g., 'Requires TRAJIR_WORKDIR or --workdir parameter').
• Add error handling guidance: 'If workdir not found, check TRAJIR_WORKDIR env var or pass --workdir explicitly. If trajectory not found, call list_nodes first to verify trajectory ID exists.'
• Document output schemas in tool descriptions or as a separate schema field. E.g., 'Returns {node_count: int, seal_count: int, by_kind: {[kind]: int}, json: bool}' so LLMs know what fields to expect.