MCP Server Dev

Braiins OS MCP Server Development

Description

Comprehensive skill for building high-quality MCP (Model Context Protocol) tools, resources, and prompts specifically for Braiins OS miner management. This skill extends the generic mcp-builder patterns with mining-specific workflows, gRPC integration, and fleet operations optimization.

Target Audience: Developers building the braiins-os-mcp-server

When to Use This Skill

Use this skill when:

• Designing new MCP tools for miner management (status queries, firmware updates, pool configuration)
• Creating MCP resources for fleet monitoring (aggregated metrics, miner status feeds)
• Building MCP prompts for guided mining operations (troubleshooting, batch updates)
• Implementing gRPC client patterns for miner communication
• Designing Redis caching strategies for fleet-scale operations
• Optimizing MCP responses for AI agent consumption

Don't use this skill for:

• General MCP server development (use mcp-builder skill instead)
• Braiins OS API reference (use braiins-os skill for documentation)
• Infrastructure deployment (use devops-related skills)
• Non-MCP related TypeScript development

Prerequisites

Knowledge Requirements

• MCP protocol fundamentals (tools, resources, prompts, transport)
• TypeScript/Node.js development
• gRPC concepts (clients, streams, error handling)
• Redis caching patterns
• Bitcoin mining operations basics

Project Context

• Codebase: braiins-os-mcp-server (TypeScript, Node.js 20.x)
• MCP SDK: @modelcontextprotocol/sdk
• gRPC: @grpc/grpc-js
• Cache: Redis 7.x
• Testing: Jest + Supertest

Related Skills

• braiins-os (.claude/skills/braiins-os/) - Braiins OS API documentation reference
• mcp-builder (docs/claude/skills-templates/mcp-builder/) - Generic MCP development guide
• CLAUDE.md (./CLAUDE.md) - Project-specific development patterns

Workflow

Phase 1: Tool Design (Mining Operations Focus)

Step 1.1: Identify Mining Workflow

Common Mining Operations:

1. Miner Status - Get current status of one or multiple miners
2. Firmware Update - Update firmware with progress tracking
3. Pool Configuration - Manage mining pool settings
4. Fleet Metrics - Aggregated statistics across all miners
5. Troubleshooting - Guided diagnostics for offline miners

Design Principle: Consolidate workflows, not just API endpoints.

Example: Firmware Update Tool Design

/**
 * ❌ BAD: Granular tools forcing agent orchestration
 */
@tool({ name: "check_firmware_version" })
async checkFirmwareVersion(minerId: string) { /* ... */ }

@tool({ name: "download_firmware" })
async downloadFirmware(url: string) { /* ... */ }

@tool({ name: "flash_firmware" })
async flashFirmware(minerId: string) { /* ... */ }

/**
 * ✅ GOOD: Consolidated workflow tool
 */
@tool({
  name: "update_miner_firmware",
  description: "Update firmware on one or more miners with automatic download, flashing, and progress tracking"
})
async updateMinerFirmware(params: {
  minerIds: string[];       // Batch operation support
  version: string;          // Target firmware version
  force?: boolean;          // Skip version checks
  progressCallback?: boolean; // Enable progress updates
}): Promise<{
  jobId: string;            // Background job ID
  status: "pending" | "running" | "completed" | "failed";
  progress: {
    total: number;
    completed: number;
    failed: number;
  };
}> {
  // Handles: version check → download → flash → verify
  // Returns job ID for async status polling
}

Step 1.2: Design Input Schema (Zod Validation)

Mining-Specific Patterns:

import { z } from "zod";

// Common validations for mining operations
const MinerIdSchema = z.string().regex(/^[a-zA-Z0-9\-_]+$/);
const MinerIdsSchema = z.array(MinerIdSchema).min(1).max(100); // Batch limit

const FirmwareVersionSchema = z.string().regex(/^\d+\.\d+\.\d+$/);

const PoolUrlSchema = z.string().url().refine(
  (url) => url.startsWith("stratum+tcp://") || url.startsWith("stratum+ssl://"),
  { message: "Pool URL must use stratum protocol" }
);

// Tool input schema
const UpdateMinerFirmwareSchema = z.object({
  minerIds: MinerIdsSchema,
  version: FirmwareVersionSchema,
  force: z.boolean().optional().default(false),
  progressCallback: z.boolean().optional().default(false)
}).strict();

Step 1.3: Design Output Format (Context-Optimized)

Principle: Agents have limited context - provide concise by default, detailed on request.

// Concise output (default) - ~150 tokens
{
  jobId: "update-20251228-abc123",
  status: "running",
  progress: {
    total: 10,
    completed: 7,
    failed: 1,
    current: "miner-008"
  },
  estimatedCompletion: "2025-12-28T18:30:00Z"
}

// Detailed output (detailLevel: "verbose") - ~500 tokens
{
  jobId: "update-20251228-abc123",
  status: "running",
  progress: {
    total: 10,
    completed: 7,
    failed: 1,
    current: "miner-008",
    breakdown: [
      { minerId: "miner-001", status: "completed", duration: "5m32s" },
      { minerId: "miner-002", status: "completed", duration: "5m28s" },
      // ... all 10 miners
    ]
  },
  startedAt: "2025-12-28T18:00:00Z",
  estimatedCompletion: "2025-12-28T18:30:00Z",
  errors: [
    {
      minerId: "miner-005",
      error: "Connection timeout",
      suggestion: "Check network connectivity with ping_miner tool"
    }
  ]
}

Phase 2: Implementation (TypeScript + MCP SDK)

Step 2.1: Tool Registration

// src/mcp/tools/updateMinerFirmware.ts
import { tool } from "@modelcontextprotocol/sdk";
import { z } from "zod";

const UpdateMinerFirmwareSchema = z.object({
  minerIds: z.array(z.string()).min(1).max(100),
  version: z.string().regex(/^\d+\.\d+\.\d+$/),
  force: z.boolean().optional().default(false),
  detailLevel: z.enum(["concise", "verbose"]).optional().default("concise")
}).strict();

@tool({
  name: "update_miner_firmware",
  description: "Update firmware on one or more Braiins OS miners. Handles download, flashing, and verification. Returns job ID for progress tracking.",
  inputSchema: UpdateMinerFirmwareSchema,
  annotations: {
    readOnlyHint: false,      // Modifies miner state
    destructiveHint: false,   // Can be rolled back
    idempotentHint: true,     // Safe to retry
    openWorldHint: true       // Interacts with external miners
  }
})
export async function updateMinerFirmware(
  params: z.infer<typeof UpdateMinerFirmwareSchema>
): Promise<FirmwareUpdateJobStatus> {
  // Implementation in next step
}

Step 2.2: gRPC Integration Pattern

import { GrpcConnectionPool } from "../../api/grpc/pool";
import { withRetry } from "../../api/grpc/retry";

export async function updateMinerFirmware(
  params: z.infer<typeof UpdateMinerFirmwareSchema>
): Promise<FirmwareUpdateJobStatus> {
  const { minerIds, version, force, detailLevel } = params;

  // 1. Create background job
  const jobId = await this.jobQueue.createJob({
    type: "firmware_update",
    minerIds,
    version
  });

  // 2. Execute updates asynchronously
  this.executeInBackground(async () => {
    const results = await Promise.allSettled(
      minerIds.map(async (minerId) => {
        // Get gRPC connection from pool
        const client = await this.grpc.pool.getConnection(minerId);

        // Update with retry logic
        return await withRetry(
          () => client.updateFirmware({ version, force }),
          { maxRetries: 3, initialDelay: 5000 }
        );
      })
    );

    // 3. Update job status
    await this.jobQueue.updateJob(jobId, {
      status: "completed",
      results: results.map((r, i) => ({
        minerId: minerIds[i],
        success: r.status === "fulfilled",
        error: r.status === "rejected" ? r.reason.message : undefined
      }))
    });
  });

  // 4. Return immediate response
  return {
    jobId,
    status: "pending",
    progress: { total: minerIds.length, completed: 0, failed: 0 }
  };
}

Step 2.3: Redis Caching Integration

Pattern: Invalidate cached data when miners update

async function updateMinerFirmware(params: /* ... */): Promise</* ... */> {
  // ... perform update ...

  // Invalidate all caches related to updated miners
  await Promise.all(
    params.minerIds.map(async (minerId) => {
      await this.redis.del(`cache:miner:${minerId}:status`);
      await this.redis.del(`cache:miner:${minerId}:config`);
    })
  );

  // Invalidate fleet-level caches
  await this.redis.del("cache:fleet:summary");

  // Publish update event for real-time subscribers
  await this.redis.publish("events:firmware-update", JSON.stringify({
    minerIds: params.minerIds,
    version: params.version,
    timestamp: new Date().toISOString()
  }));

  return { /* ... */ };
}

Phase 3: Resource Development (Fleet Monitoring)

Step 3.1: Design Resource URI Scheme

URI Pattern: braiins:///<category>/<resource>[/<identifier>]

braiins:///fleet/summary             # Aggregated fleet metrics
braiins:///fleet/miners              # List of all miners
braiins:///miner/miner-123/status    # Single miner status
braiins:///miner/miner-123/logs      # Miner logs
braiins:///jobs/update-abc123        # Job status

Step 3.2: Implement Cached Resource

import { resource } from "@modelcontextprotocol/sdk";

@resource({
  uri: "braiins:///fleet/summary",
  name: "Fleet Summary",
  description: "Aggregated metrics for all managed miners (cached for 30s)",
  mimeType: "application/json"
})
export async function getFleetSummary(): Promise<FleetSummary> {
  const cacheKey = "cache:fleet:summary";
  const ttl = 30; // 30 seconds - fleet data changes slowly

  // Check cache
  const cached = await this.redis.get(cacheKey);
  if (cached) {
    return JSON.parse(cached);
  }

  // Compute fresh data (expensive operation)
  const [total, online, hashrate, temp, alerts] = await Promise.all([
    this.db.countMiners(),
    this.db.countMinersOnline(),
    this.aggregateHashrate(),       // Sum across all miners
    this.averageTemperature(),      // Average temp
    this.getActiveAlerts()          // Count critical alerts
  ]);

  const summary: FleetSummary = {
    totalMiners: total,
    onlineMiners: online,
    offlineMiners: total - online,
    totalHashrate: `${hashrate} TH/s`,
    averageTemp: `${temp}°C`,
    activeAlerts: alerts.length,
    lastUpdated: new Date().toISOString()
  };

  // Cache for 30 seconds
  await this.redis.setex(cacheKey, ttl, JSON.stringify(summary));

  return summary;
}

Phase 4: Prompt Development (Guided Workflows)

Step 4.1: Troubleshooting Prompt

import { prompt } from "@modelcontextprotocol/sdk";

@prompt({
  name: "troubleshoot_miner_offline",
  description: "Step-by-step guide to diagnose and fix offline miners",
  arguments: [
    { name: "minerId", description: "Miner ID to troubleshoot", required: true }
  ]
})
export async function troubleshootMinerOffline(
  minerId: string
): Promise<PromptMessage[]> {
  // Get miner details
  const miner = await this.db.getMiner(minerId);

  return [
    {
      role: "user",
      content: {
        type: "text",
        text: `My miner ${minerId} is offline. Can you help me fix it?`
      }
    },
    {
      role: "assistant",
      content: {
        type: "text",
        text: `I'll help troubleshoot miner ${minerId}. Let me run through diagnostics:

**Step 1: Check Last Known Status**
Last seen: ${miner.lastSeen || "Never"}
Last status: ${miner.lastStatus || "Unknown"}
Last IP: ${miner.lastIp || "Unknown"}

**Step 2: Test Network Connectivity**
I'll ping the miner to check if it's reachable.
→ Using tool: ping_miner({ minerId: "${minerId}" })

Please wait while I test connectivity...

**Next Steps Based on Results:**
✅ If ping succeeds:
   - Use tool: get_miner_logs({ minerId: "${minerId}", lines: 50 })
   - Check for error patterns in logs
   - May need to restart miner services

❌ If ping fails:
   - Check physical power connection
   - Verify network cable connected
   - Check router/switch port status
   - Verify IP address hasn't changed

Would you like me to proceed with the ping test?`
      }
    }
  ];
}

Phase 5: Testing (MCP-Specific)

Step 5.1: Unit Test for MCP Tool

// tests/unit/mcp/tools/updateMinerFirmware.test.ts
import { MCPServer } from "../../../../src/server";
import { createMockGrpcClient } from "../../../mocks/grpc";
import { createMockRedis } from "../../../mocks/redis";

describe("update_miner_firmware tool", () => {
  let mcpServer: MCPServer;
  let mockGrpc: jest.Mocked<GrpcClient>;
  let mockRedis: jest.Mocked<Redis>;

  beforeEach(() => {
    mockGrpc = createMockGrpcClient();
    mockRedis = createMockRedis();
    mcpServer = new MCPServer({ grpc: mockGrpc, redis: mockRedis });
  });

  it("should accept valid firmware update request", async () => {
    const result = await mcpServer.callTool("update_miner_firmware", {
      minerIds: ["miner-1", "miner-2"],
      version: "2.0.1"
    });

    expect(result.jobId).toBeDefined();
    expect(result.status).toBe("pending");
    expect(result.progress.total).toBe(2);
  });

  it("should reject invalid version format", async () => {
    await expect(
      mcpServer.callTool("update_miner_firmware", {
        minerIds: ["miner-1"],
        version: "invalid-version"
      })
    ).rejects.toThrow("version must match format");
  });

  it("should enforce batch size limit", async () => {
    const tooManyMiners = Array.from({ length: 101 }, (_, i) => `miner-${i}`);

    await expect(
      mcpServer.callTool("update_miner_firmware", {
        minerIds: tooManyMiners,
        version: "2.0.1"
      })
    ).rejects.toThrow("maximum 100 miners");
  });
});

Step 5.2: Integration Test for Resource

// tests/integration/mcp/resources/fleetSummary.test.ts
describe("Fleet Summary Resource", () => {
  let mcpServer: MCPServer;
  let redis: Redis;

  beforeAll(async () => {
    redis = new Redis(process.env.REDIS_URL);
    mcpServer = new MCPServer({ redis });
  });

  it("should return cached fleet summary on second request", async () => {
    const result1 = await mcpServer.readResource("braiins:///fleet/summary");
    const result2 = await mcpServer.readResource("braiins:///fleet/summary");

    expect(result1).toEqual(result2);

    // Verify cache was used (mock/spy on Redis get method)
    expect(redis.get).toHaveBeenCalledWith("cache:fleet:summary");
  });

  it("should refresh cache after TTL expiry", async () => {
    await mcpServer.readResource("braiins:///fleet/summary");

    // Wait for cache to expire (30s TTL)
    await sleep(31000);

    const result = await mcpServer.readResource("braiins:///fleet/summary");

    expect(result.lastUpdated).not.toBe(/* previous timestamp */);
  });
});

Mining-Specific Patterns

Pattern 1: Batch Operations with Progress Tracking

Use Case: Update firmware on 50 miners simultaneously

@tool({ name: "update_multiple_miners" })
async updateMultipleMiners(params: {
  minerIds: string[];
  operation: "firmware_update" | "pool_change" | "reboot";
  config: any;
}) {
  // Create job for background execution
  const jobId = uuid();

  // Process in parallel with concurrency limit
  const concurrency = 10; // Max 10 simultaneous updates
  const queue = new PQueue({ concurrency });

  const promises = params.minerIds.map((minerId) =>
    queue.add(async () => {
      try {
        await this.performOperation(minerId, params.operation, params.config);
        await this.updateJobProgress(jobId, { completed: minerId });
      } catch (error) {
        await this.updateJobProgress(jobId, { failed: minerId, error });
      }
    })
  );

  // Don't wait - return job ID immediately
  Promise.all(promises).then(() => {
    this.updateJobStatus(jobId, "completed");
  });

  return {
    jobId,
    status: "running",
    total: params.minerIds.length,
    pollUrl: `braiins:///jobs/${jobId}`
  };
}

Pattern 2: Real-Time Status Streaming

Use Case: Subscribe to miner status updates via Redis pub/sub

@tool({ name: "subscribe_miner_status" })
async subscribeMinerStatus(params: { minerId: string }) {
  // Start gRPC stream
  const stream = await this.grpc.streamMinerStatus(params.minerId);

  // Publish to Redis for agent consumption
  for await (const status of stream) {
    await this.redis.publish(
      `miner:${params.minerId}:status`,
      JSON.stringify(status)
    );
  }

  return {
    subscribed: true,
    channel: `miner:${params.minerId}:status`,
    message: "Status updates will be published to Redis pub/sub channel"
  };
}

// Agent can then read from resource:
@resource({ uri: "braiins:///miner/{minerId}/status/stream" })
async getMinerStatusStream(minerId: string) {
  // Subscribe to Redis channel and return latest status
  const status = await this.redis.get(`miner:${minerId}:status:latest`);
  return JSON.parse(status);
}

Pattern 3: Actionable Error Guidance

Use Case: Miner unreachable - guide agent to next steps

async function pingMiner(minerId: string) {
  try {
    const client = await this.grpc.pool.getConnection(minerId);
    await client.ping({ timeout: 5000 });

    return { reachable: true, latency: "45ms" };
  } catch (error) {
    // Return actionable error with next steps
    return {
      reachable: false,
      error: {
        code: "MINER_UNREACHABLE",
        message: `Cannot connect to miner ${minerId}`,
        suggestions: [
          "Check if miner is powered on",
          "Verify network connectivity with: list_miners",
          "Check miner IP address in configuration",
          "Try rebooting miner with: reboot_miner"
        ],
        possibleCauses: [
          "Miner offline or powered off",
          "Network firewall blocking gRPC port (50051)",
          "Incorrect IP address in database",
          "Miner experiencing hardware failure"
        ]
      }
    };
  }
}

Quality Standards

MCP Tool Checklist

• Input Validation: Zod schema with proper constraints
• Batch Support: Handles multiple miners when applicable
• Concise Output: Default response < 300 tokens
• Detailed Option: Verbose mode available via detailLevel param
• Error Guidance: Errors include suggestions for next steps
• Caching: Reads use cached data when appropriate
• Cache Invalidation: Writes invalidate related caches
• Background Jobs: Long operations return job ID immediately
• Progress Tracking: Job progress available via separate tool/resource
• Annotations: readOnlyHint, destructiveHint, idempotentHint, openWorldHint set correctly
• Documentation: Clear description with examples
• Tests: Unit tests cover happy path + error cases

MCP Resource Checklist

• URI Format: Follows braiins:///<category>/<resource> pattern
• Caching: Appropriate TTL for data type
• Freshness: lastUpdated timestamp included
• Mime Type: Correct content type (application/json, text/plain, etc.)
• Performance: Queries optimized for large fleets
• Tests: Integration tests verify caching behavior

MCP Prompt Checklist

• Guided Workflow: Clear step-by-step instructions
• Tool References: Suggests specific tools for each step
• Resource References: Links to relevant resources
• Conditional Logic: Different paths based on outcomes
• Next Steps: Always provides clear next actions
• Context-Aware: Uses miner-specific data in guidance

Common Pitfalls

❌ Pitfall 1: Not Supporting Batch Operations

Problem: Tool only accepts single miner ID

// BAD
@tool({ name: "get_miner_status" })
async getMinerStatus(minerId: string) { /* ... */ }

// Agent must call 50 times for 50 miners

Solution: Accept array of IDs

// GOOD
@tool({ name: "get_miner_status" })
async getMinerStatus(minerIds: string[]) {
  return Promise.all(minerIds.map(id => this.fetchStatus(id)));
}

// Agent calls once for all 50 miners

❌ Pitfall 2: Returning Too Much Data

Problem: Tool returns 5000+ token response overwhelming context

// BAD - Returns full miner details
{
  minerId: "miner-123",
  status: "running",
  hashrate: { /* 20 fields */ },
  temperature: { /* 15 fields */ },
  fans: [ /* 6 fans × 10 fields each */ ],
  pools: [ /* 3 pools × 30 fields each */ ],
  // ... 4500 more tokens
}

Solution: Concise by default, detailed on request

// GOOD - Concise default
{
  minerId: "miner-123",
  status: "running",
  hashrate: "95 TH/s",
  temp: "65°C",
  issues: []  // Only show if problems exist
}

// Detailed available via detailLevel: "verbose"

❌ Pitfall 3: Blocking on Long Operations

Problem: Firmware update takes 10 minutes, blocking agent context

// BAD
@tool({ name: "update_firmware" })
async updateFirmware(params: {/* ... */}) {
  // Wait for 10 minute update to complete
  await this.performUpdate(params);

  return { status: "completed" };
}
// Agent stuck for 10 minutes

Solution: Return job ID immediately, poll separately

// GOOD
@tool({ name: "update_firmware" })
async updateFirmware(params: {/* ... */}) {
  const jobId = await this.startBackgroundUpdate(params);

  return {
    jobId,
    status: "pending",
    pollWith: "check_job_status"
  };
}

// Separate tool for polling
@tool({ name: "check_job_status" })
async checkJobStatus(jobId: string) {
  return await this.jobQueue.getStatus(jobId);
}

Integration Notes

Related Skills

• braiins-os - Braiins OS API documentation and reference
• mcp-builder - Generic MCP server development patterns
• grpc-client-dev (planned) - gRPC client implementation patterns
• redis-caching-patterns (planned) - Advanced Redis caching strategies

Related Commands

• /start-session - Initialize development session with project context
• /close-session - End session with documentation updates
• /test-all - Run comprehensive test suite
• /test-mcp-tools (planned) - Test MCP tools in isolation

Related Agents

• Architect - System design and API design review
• Builder - Feature implementation
• Validator - Testing and code review
• Scribe - Documentation updates

Version History

• 1.0.0 (2025-12-28): Initial skill creation
  • Mining-specific MCP development patterns
  • Tool/Resource/Prompt workflows
  • gRPC integration patterns
  • Redis caching strategies
  • Batch operations and progress tracking
  • Quality checklists and common pitfalls

Metadata

{
  "name": "mcp-server-dev",
  "version": "1.0.0",
  "description": "Braiins OS MCP Server Development Skill",
  "author": "braiins-os-mcp-server project",
  "created": "2025-12-28",
  "status": "active",
  "complexity": "complex",
  "category": "mcp-development",
  "tags": ["mcp", "braiins-os", "mining", "grpc", "redis", "typescript"],
  "extends": "mcp-builder",
  "related_skills": ["braiins-os", "mcp-builder"],
  "target_project": "braiins-os-mcp-server"
}