Both tools have severely incomplete definitions. Tool schemas are present but minimal, input parameters lack type definitions, descriptions are vague and under 20 chars, and output schemas are entirely undocumented. No parameter descriptions exist. Tool names use a namespace prefix (boundary/) but lack clear action verbs. Error handling, validation guidance, and recovery paths are absent. This server does not meet baseline quality for production agent use.

• Expand tool descriptions to 100 - 200 characters. Explain: (1) what the tool does, (2) when to call it vs. the other tool, (3) what it returns. Example: 'Start a boundary session with an intent envelope declaring authorized capabilities, sources, and limits. Call this once per agent session before evaluating actions. Returns session_id and policy_context.'
• Add full JSON Schema to input parameters. For 'intent_envelope', specify: type='object', required=['capabilities', 'sources'], properties with type and description for each field (e.g., capabilities: array of strings, sources: array of objects with id/trust/endpoint).
• Document output schemas for both tools. Specify return type (object), all fields (session_id, policy_context, evaluation_result, etc.), their types, and whether they are required. Include example values.
• Add parameter descriptions. For 'intent_envelope': 'Object containing authorized capabilities (array of strings), sources (array of {id, trust, endpoint}), and limits (object with max_requests, max_age_hours). Required.' For 'action': 'Object with type (string), capability (string), endpoint (string), and payload (object). Evaluated against active session policy.'
• Add error handling guidance. Document: (1) what errors can occur (e.g., 'session not found', 'policy violation'), (2) whether they are retryable, (3) what the agent should do (e.g., 'Call boundary/session.start first', 'Check intent_envelope format').

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained