mcp-builder

MCP Builder — Context Protocol Mastery

Mandatory Pre-Flight Context Inspection

Before reading, generating, or refactoring code in the mcp-builder domain, inspect these 5 critical parameters:

1. System Boundaries & Dependencies: Verify that all required dependencies exist in target package manifests and environment paths.
2. Runtime Context & Platform Invariants: Confirm target platform constraints (Node.js, Browser, Mobile OS, Edge runtime) before applying APIs.
3. Execution Guardrails: Identify potential side-effects, state mutations, and unhandled asynchronous exceptions.
4. Validation & Type Contracts: Validate input data schemas and strict type constraints across all module interfaces.
5. Observability & Proof of Execution: Ensure execution produces tangible verification signals (terminal output, tests, metrics).

Activation Boundaries

• Activate when: Use when executing, coordinating, planning, or reviewing mcp builder agent workflows, cognitive loops, and architecture standards.
• DO NOT activate when: The task falls outside the mcp-builder domain or is managed by a different dedicated specialist agent.

🔁 Multi-Pass Execution Protocol

🛠️ Technical Architecture & Reference Recipes

Hallucination Traps (Read First)

• ❌ Exposing tools without input validation schemas -> ✅ Every MCP tool MUST have JSON Schema for parameters; the protocol requires it
• ❌ Returning unstructured strings from tool calls -> ✅ Return structured JSON that the LLM can reliably parse and act on
• ❌ Not handling tool call timeouts -> ✅ Always set execution timeouts; hanging tools block the entire LLM conversation loop

1. The Anatomy of an MCP Server

The Model Context Protocol (MCP) standardizes how AI agents fetch local data and execute tools. A robust MCP server exposes exactly 3 primary concepts:

1. Resources: Read-only data payloads (Logs, local files, database dumps).
2. Prompts: Reusable injected context scaffolding (e.g., "Summarize this log with strict parameters").
3. Tools: Actionable executed capabilities (e.g., "Run Postgres Query", "Restart Server").

// Standardize exposing a Tool securely via an MCP Server Wrapper
import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
import { z } from 'zod';

const server = new McpServer({
  name: 'internal-database-auditor',
  version: '1.0.0',
});

// Defining a rigorous tool parameter boundary
server.tool(
  'query_production_database',
  'Executes a read-only sanitized query against the production analytical replica.',
  {
    table: z
      .enum(['users', 'transactions', 'audit_logs'])
      .describe('The specific table to analyze'),
    limit: z.number().max(100).default(10).describe('Maximum row returns to prevent context bloat'),
  },
  async ({ table, limit }) => {
    // Execution logic
    const data = await secureDatabaseClient.query(`SELECT * FROM ${table} LIMIT ${limit}`);
    return {
      content: [{ type: 'text', text: JSON.stringify(data) }],
    };
  },
);

2. Resource Management vs Tool Management

Do not use a Tool to read static data. Do not use a Resource to invoke remote actions.

• Resources (URI based): Act identically to local files. Exposed explicitly so the AI context manager can read them before invoking tools. Use for things like file:///app/config.json or db://schema/users.
• Tools: Use exclusively when parameterized execution is required dynamically. Tools MUST be accompanied by extremely literal, explicit descriptions, because the LLM uses the description text to map Intent to the Tool execution.

3. Structuring Tool Descriptions (The LLM Gateway)

The LLM decides to fire your tool based entirely on the Description schema. If your description is vague, the LLM will hallucinate executions unpredictably.

// ❌ VAGUE (The LLM will guess when to use this, often incorrectly)
description: 'Changes the system status.';

// ✅ DETERMINISTIC (The LLM knows the exact boundaries and consequences)
description: "Transitions the payment processing gateway between 'ACTIVE' and 'MAINTENANCE' modes. Use this ONLY after verifying traffic logs to halt impending queue flooding. Requires Admin clearance.";

4. MCP Security Boundaries

An MCP Server gives an external AI execution capability over your shell or database.

• Never Expose Raw Shells Natively: Unless deliberately building a high-trust local desktop agent. Expose mapped commands (execute_npm_build) instead of raw terminals (bash_command).
• Enforce Read-Only Defaults: If creating a database tool, create query_select_only separate from execute_mutation. Give the AI read-only access.
• Context Size Truncation: If a tool queries a 5GB text log, the AI context window will instantly overflow and crash the session. The MCP logic MUST forcibly truncate outputs before returning.

🚨 Edge-Case & Failure Mode Matrix

🏛️ Tribunal Verification & Guardrails

Active Reviewers: orchestrator · agent-organizer · logic-reviewer Slash Command: /review or /tribunal-full

🔬 Evidence Standard (Tri-State Verification)

Every finding, audit statement, or completion claim must classify its factual certainty:

• [OBSERVED]: Directly confirmed in the codebase or verified via executed terminal command.
• [INFERRED]: Logically deduced from code patterns, architectural data flow, or schema relations.
• [UNVERIFIED]: Speculative hypothesis or runtime possibility requiring active testing or measurement.

✅ Pre-Flight Self-Audit Checklist

✅ Did I deconstruct the root objective before proposing architecture?
✅ Did I identify dependencies, bottlenecks, and parallelizable sub-tasks?
✅ Did I avoid over-engineering and select the simplest effective pattern?
✅ Did I verify assumptions with concrete file reads instead of speculation?
✅ Did I establish measurable verification criteria before completion?

🛑 Verification-Before-Completion (VBC) Protocol

CRITICAL: You must follow a strict "evidence-based closeout" state machine.

• ❌ Forbidden: Declaring a task complete because the output "looks correct."
• ✅ Required: You are explicitly forbidden from finalizing any task without providing concrete evidence (terminal output, passing test suites, compiler success, or equivalent operational proof) that your output works as intended.