Usage Examples

Real-world scenarios demonstrating effective use of the MCP plugin for documentation access, semantic code analysis, and custom MCP server development.

Examples

Setting Up Documentation for a New Project

Scenario: You're starting a new Next.js project with Prisma and want up-to-date documentation access from the start.

# Set up the project
> npx create-next-app@latest my-app

# Install the MCP plugin
> /plugin install mcp@NeoLabHQ/context-engineering-kit

# Configure documentation access
> /mcp:setup-context7-mcp nextjs 14, prisma, typescript, zod

Expected Flow:

1. Command checks if Context7 MCP is already configured

2. If not, guides you through installation for your environment

3. Searches Context7 for documentation IDs for each technology

4. Updates CLAUDE.md with recommended library IDs

Result in CLAUDE.md:

### Use Context7 MCP for Loading Documentation

Context7 MCP is available to fetch up-to-date documentation with code examples.

**Recommended library IDs**:

- `nextjs` - Next.js 14 documentation with App Router
- `prisma` - Prisma ORM schema and client documentation
- `typescript` - TypeScript language reference
- `zod` - Zod schema validation library

Using the Documentation:

After setup, when you ask about Next.js Server Actions or Prisma relations, the LLM queries Context7 for current documentation instead of relying on potentially outdated training data.

---

Auto-Detecting Project Technologies

Scenario: You've inherited a project and want to set up documentation access without manually listing all dependencies.

# Navigate to project
> cd existing-project

# Let the command analyze your project
> /mcp:setup-context7-mcp

Expected Flow:

1. Command scans package.json, requirements.txt, or other dependency files

2. Identifies React, Redux, Express, and PostgreSQL

3. Searches Context7 for matching documentation

4. Updates CLAUDE.md with discovered technologies

When Analysis Finds Technologies:

Analyzed project structure:
- Found package.json with React 18, Redux Toolkit, Express
- Found docker-compose.yml with PostgreSQL
- Detected TypeScript configuration

Setting up documentation for: react, @reduxjs/toolkit, express, postgresql, typescript

---

Setting Up Semantic Code Analysis for a Large Codebase

Scenario: You're working on a large monorepo and need better code navigation than grep can provide.

# Set up Serena for semantic analysis
> /mcp:setup-serena-mcp

Expected Flow:

1. Command checks for Serena availability

2. Verifies uv is installed

3. Detects your MCP client (Claude Code in this case)

4. Guides through configuration if needed

5. Initializes project indexing

6. Updates CLAUDE.md with usage guidelines

Setup Interaction:

Checking Serena MCP availability...
- Serena tools not accessible

Prerequisites check:
- uv: Installed (0.4.18)

Detected client: Claude Code

Configuration needed:
1. Add Serena to your MCP configuration
2. Restart Claude to load the new MCP server

After restart, run: /mcp:setup-serena-mcp continue

After Successful Setup:

Serena MCP is now configured!

Running project indexing...
- Indexed 1,247 files
- Found 8,432 symbols

Testing connection:
- find_symbol: Working
- list_symbols: Working
- read_file: Working

Updated CLAUDE.md with Serena usage guidelines.

Using Semantic Operations:

# Find a specific function across the codebase
> "Find all usages of the validateUser function"

# Serena uses semantic search instead of text grep
# Returns: function definition, all call sites, and type information

---

Combining Context7 and Serena for Maximum Effectiveness

Scenario: You want both documentation access and semantic code understanding for a complex project.

# First, set up documentation access
> /mcp:setup-context7-mcp react, graphql, @apollo/client

# Then, set up semantic code analysis
> /mcp:setup-serena-mcp

Result in CLAUDE.md:

### Use Context7 MCP for Loading Documentation

Context7 MCP is available to fetch up-to-date documentation with code examples.

**Recommended library IDs**:

- `react` - React 18 documentation
- `graphql` - GraphQL specification and best practices
- `@apollo/client` - Apollo Client documentation

### Use Serena MCP for Semantic Code Analysis

Serena MCP is available for advanced code retrieval and editing capabilities.

- Use Serena's tools for precise code manipulation in structured codebases
- Prefer symbol-based operations over file-based grep/sed operations

Key usage points:
- Use `find_symbol` to locate functions, classes, and types by name
- Use `list_symbols` to explore available symbols in a file or module
- Prefer semantic operations for refactoring over text replacement

Workflow Benefits:

1. When implementing a new GraphQL resolver, Context7 provides current Apollo Client patterns

2. When refactoring, Serena finds all symbol usages across the codebase

3. Both tools work together: documentation for "how to do it", semantic analysis for "where to do it"

---

Building a Custom MCP Server for Slack Integration

Scenario: Your team uses Slack extensively and you want the LLM to interact with Slack channels and messages.

# Start the MCP server development guide
> /mcp:build-mcp

# When prompted, describe your use case:
> "I want to build an MCP server that lets the LLM read and post messages
   to Slack channels, search message history, and manage channel membership"

Phase 1 Output - Research and Planning:

## Implementation Plan: Slack MCP Server

### Tool Selection (Priority Order)

1. **search_messages** - Search across channels with filters
   - High value: Enables finding relevant context quickly
   - Workflow: Single tool replaces manual channel browsing

2. **post_message** - Send messages to channels
   - Essential for communication workflows
   - Include thread support for context

3. **list_channels** - Get available channels
   - Foundation for other operations
   - Support filtering by membership/type

4. **get_channel_history** - Read recent messages
   - Support for pagination
   - Include thread expansion option

5. **manage_channel_membership** - Join/leave channels
   - Lower priority but enables autonomy

### Input/Output Design

- Use "concise" vs "detailed" response formats
- Limit message returns to 25 by default
- Support pagination for longer histories
- Return human-readable timestamps, not Unix epochs

Phase 2 - Implementation Example:

# slack_mcp_server.py (generated with guidance)
from mcp import Server, Tool
from pydantic import BaseModel, Field

class SearchMessagesInput(BaseModel):
    """Search parameters for Slack messages."""
    query: str = Field(
        description="Search query. Supports Slack search operators.",
        examples=["from:@alice deployment", "in:#engineering bug"]
    )
    channels: list[str] | None = Field(
        default=None,
        description="Limit search to specific channels. Omit to search all."
    )
    limit: int = Field(
        default=10,
        ge=1,
        le=100,
        description="Maximum messages to return. Default 10 for context efficiency."
    )
    format: str = Field(
        default="concise",
        pattern="^(concise|detailed)$",
        description="concise: message + author + timestamp. detailed: includes reactions, threads."
    )

@server.tool
async def search_messages(params: SearchMessagesInput) -> str:
    """
    Search Slack messages across channels.

    Use this tool to find relevant conversations, decisions, or context.
    Returns messages sorted by relevance with author and timestamp.

    Example queries:
    - "deployment friday" - Find messages about Friday deployments
    - "from:@alice in:#engineering" - Alice's messages in #engineering
    - "has:reaction" - Messages with reactions (indicates importance)

    For actionable results, combine with post_message to respond in context.
    """
    # Implementation...

Phase 4 - Evaluation Questions:

<evaluation>
  <qa_pair>
    <question>Search for messages containing "deployment" in the #engineering channel from the last week. Who posted the most recent one?</question>
    <answer>alice</answer>
  </qa_pair>
  <qa_pair>
    <question>Find the channel with the most members that discusses backend topics. What is its name?</question>
    <answer>backend-general</answer>
  </qa_pair>
</evaluation>

---

Building a Python MCP Server for Internal API

Scenario: Your company has an internal API for managing customer data and you want LLM access.

> /mcp:build-mcp

> "Build an MCP server for our internal Customer API at https://api.internal.com
   I need to search customers, view customer details, and update customer status"

Design Principles Applied:

## Tool Design for Customer API

### Workflow-Oriented Design

**Instead of**:
- get_customer_by_id
- get_customer_by_email
- get_customer_by_name

**We build**:
- search_customers (handles all lookup methods with smart defaults)

### Context-Optimized Responses

**Instead of**:
```json
{
  "id": "cust_123",
  "created_at": 1699574400,
  "updated_at": 1699660800,
  "metadata": {...},
  "billing_info": {...},
  ...100 more fields
}

We return:

Customer: John Doe (cust_123)
Email: [email protected]
Status: Active | Since: Nov 2024
Recent Activity: 3 orders in last 30 days

[Use format="detailed" for full record]

Actionable Errors

Instead of:

Error: 404 Not Found

We return:

Customer not found with ID 'cust_123'.
Try:
- search_customers with email or name
- Check if ID format is correct (should be 'cust_' prefix)

---

### Using MCP in a Feature Development Workflow

**Scenario**: Complete feature development using MCP servers alongside other plugins.

```bash
# Set up MCP servers at project start
> /mcp:setup-context7-mcp react, typescript, prisma
> /mcp:setup-serena-mcp

# Research phase - Context7 provides current documentation
> /sdd:01-research

# Implementation - Serena helps navigate codebase
> /sdd:04-implement

# Review with semantic understanding
> /reflexion:critique

# Save learnings
> /reflexion:memorize

How MCP Enhances Each Phase:

Phase	MCP Contribution
Research	Context7 provides current API documentation
Implementation	Serena finds related code and symbols
Review	Serena enables precise code navigation during critique
Documentation	Context7 ensures examples match current APIs

---

Troubleshooting: Context7 Documentation Not Found

Scenario: Context7 cannot find documentation for a library you need.

> /mcp:setup-context7-mcp obscure-library

# Response:
# Documentation not found for: obscure-library
#
# Suggestions:
# 1. Check library name spelling
# 2. Try alternative names (e.g., 'react-query' vs '@tanstack/react-query')
# 3. Library may not be indexed - consider contributing to Context7
# 4. Use WebFetch to load documentation directly from library repository

Workaround:

# Manually add documentation source to CLAUDE.md
> "Add a note to CLAUDE.md that for obscure-library documentation,
   load from https://github.com/org/obscure-library/blob/main/docs/api.md"

---

Troubleshooting: Serena Indexing Failures

Scenario: Serena fails to index your project correctly.

> /mcp:setup-serena-mcp

# Issue: Indexing failed for src/generated/*
# These files are auto-generated and should be excluded

Resolution:

# Configure Serena to exclude generated files
> "Configure Serena to exclude the src/generated directory from indexing"

# Re-run indexing
> /mcp:setup-serena-mcp

# Result:
# Indexing complete:
# - Excluded: src/generated/* (1,234 files)
# - Indexed: 456 files
# - Found: 2,341 symbols

Integration with Other Plugins

With Reflexion

# Set up documentation, then verify configuration
> /mcp:setup-context7-mcp react, nextjs
> /reflexion:reflect

# After building a custom MCP server, get comprehensive review
> /mcp:build-mcp
> /reflexion:critique
> /reflexion:memorize "MCP server design patterns"

With SDD (Spec-Driven Development)

# Documentation access improves research quality
> /mcp:setup-context7-mcp
> /sdd:01-research

# Semantic analysis helps implementation
> /mcp:setup-serena-mcp
> /sdd:04-implement

With TDD

# Semantic analysis helps find test targets
> /mcp:setup-serena-mcp
> /tdd:write-tests

With Kaizen

# Semantic analysis aids root cause investigation
> /mcp:setup-serena-mcp
> /kaizen:why "Why is the authentication failing?"