MCP Integration

90+ Tools for AI Assistants

Give your AI coding assistant deep code understanding via the Model Context Protocol. Works with Claude Code, Cursor, Windsurf, VS Code, and more.

What is MCP?

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools. Instead of your AI grepping through files, it can query CKB's semantic understanding of your codebase.

This means your AI knows:

- Who calls a function (not just where it's mentioned)
- What breaks if you change something
- Who should review code changes
- Whether code is actually used in production

# Without CKB
AI: "I found 47 matches for Handler"

# With CKB
AI: "HandleRequest is called by 3 routes:
     - POST /api/checkout via CheckoutHandler
     - POST /api/refund via RefundHandler
     - POST /api/webhook via WebhookHandler

     Changing it affects 12 files.
     Risk score: 73 (High)
     Suggested reviewers: @alice, @bob"

83% Token Savings

Most sessions only need a subset of tools. Presets reduce token overhead dramatically while letting you expand mid-session if needed.

Preset	Tools	Tokens	Use Case
`core`	21	~2k	Essential navigation, impact, compound ops
`review`	27	~3k	Core + diff, ownership, reviewers
`refactor`	33	~4k	Core + cycles, coupling, dead code, suggestions
`docs`	27	~3k	Core + doc-symbol linking
`ops`	33	~4k	Core + jobs, webhooks, metrics
`federation`	35	~4k	Core + cross-repo queries
`full`	92	~11k	All tools

# Default: core preset (21 tools, ~2k tokens)
ckb mcp

# For code review sessions
ckb mcp --preset=review

# AI can expand mid-session
> "I need to check cross-repo dependencies"
AI: [calls expandToolset with federation preset]

Tool Categories

Navigation

searchSymbolsgetSymbolfindReferencesgetCallGraphtraceUsagelistEntrypoints

Impact

analyzeImpactprepareChangegetAffectedTestscompareAPIgetHotspotsfindCycles

Understanding

explainSymbolexplainFileexplainPathlistKeyConceptsgetArchitecture

Ownership

getOwnershipgetReviewersgetOwnershipDriftgetModuleResponsibilities

Plus 50+ more tools for security, documentation, federation, telemetry, and automation.

v8.0 Enhancements

Version 8.0 brought major improvements to the MCP experience.

Compound Operations

One call replaces 5. explore, understand, and prepareChange aggregate multiple tools.

Batch Operations

batchGet (50 symbols) and batchSearch (10 queries) in single calls.

Streaming Results

Real-time progress with Server-Sent Events for long operations.

Structured Errors

6 error types with remediation guidance. No more cryptic failures.

Confidence Scoring

Every result includes confidence tier (high/medium/low) with reasons.

Cache Transparency

See if results came from cache. Track hit rates in status.

v8.1 Refactoring Intelligence

New tools for proactive refactoring: cycle detection, move/relocate support, variable flow analysis, and automated suggestions.

Dependency Cycle Detection

findCycles uses Tarjan's SCC to detect circular dependencies and recommends which edge to break.

Move/Relocate Support

prepareChange with changeType "move" tracks affected imports and detects target conflicts.

Extract Flow Analysis

Tree-sitter-based variable flow analysis identifies parameters, returns, and locals for extract refactoring.

Refactoring Suggestions

suggestRefactorings proactively detects opportunities: complex functions, coupling, dead code, test gaps.

Unified Refactor Planning

planRefactor combines impact analysis with step-by-step refactoring plan generation.

Test Gap Analysis

analyzeTestGaps finds high-risk code that lacks test coverage.

Quick Setup

Claude Code

# Auto-configure
ckb setup

# Or manually add to .mcp.json
{
  "mcpServers": {
    "ckb": {
      "command": "npx",
      "args": ["@tastehub/ckb", "mcp"]
    }
  }
}

Cursor / Windsurf / VS Code

# Cursor: ~/.cursor/mcp.json
# Windsurf: ~/.codeium/windsurf/mcp_config.json
# VS Code: settings.json

{
  "mcpServers": {
    "ckb": {
      "command": "npx",
      "args": ["@tastehub/ckb", "mcp", "--preset=review"]
    }
  }
}

Example Prompts

"What calls the ProcessPayment function?"

AI uses getCallGraph to show callers with full paths

"What breaks if I change UserService?"

AI uses prepareChange for blast radius + affected tests

"Who should review this PR?"

AI uses getReviewers with ownership data

"Is this function still used in production?"

AI uses findDeadCode + getObservedUsage

Supercharge your AI

Give your coding assistant the context it needs. Setup takes 60 seconds.

MCP Documentation Prompt Cookbook