CrewAI MCP Dev Tutorial Agents

A multi-agent system that generates technical integration tutorials by analyzing real source code, scanning for vulnerabilities, and writing production-quality guides — powered by CrewAI, Google Gemini, and 11 production MCP servers from the Vinkius AI Gateway.

---

Why Most Developer Tutorials Fail

Developers constantly search for integration guides — "How to integrate Stripe with Supabase in Next.js", "Connect Redis to FastAPI with Docker", "Set up Sentry in a Django project". These long-tail technical queries have low competition but attract highly qualified traffic: developers actively building something.

The problem is that most tutorials available today are written against outdated versions, use deprecated packages, skip error handling, and omit critical configuration steps. The author wrote the guide six months ago, the SDK had a breaking change in v3.2, and now hundreds of developers are hitting the same cryptic error on step 7.

This project takes a fundamentally different approach. Instead of writing from memory, three AI agents read the actual source code from GitHub, validate every dependency against vulnerability databases, and then write the tutorial with verified, current code. The critical enabler is the Model Context Protocol (MCP) — an open standard that gives AI agents direct access to developer platforms through a unified interface.

---

How It Works

The system operates as a sequential pipeline of three specialized agents, each connected to MCP servers hosted on the Vinkius AI Gateway.

Phase 1 — Source Code Analysis

The first agent acts as a senior engineer performing a code review of the integration. It connects to four MCP servers:

• GitHub to read repository code, browse file structures, check releases, and find official integration examples
• GitLab for projects hosted on GitLab, including CI/CD configurations
• Docker Hub to identify official images, recommended tags, and deployment configurations
• Firecrawl to scrape the official documentation pages for both tools, extracting the latest API references and setup guides

The agent reads the actual source code before suggesting any integration pattern. If the library changed its authentication method in the latest release, the agent knows — because it just read the code.

Phase 2 — Code Review and Validation

The second agent reviews everything the first agent proposed. It connects to five MCP servers:

• Snyk to scan every dependency for known security vulnerabilities
• SonarCloud to verify the code follows quality best practices
• Sentry to check common runtime errors other developers have encountered with this integration
• SwaggerHub to validate API specifications and ensure the proposed endpoints are correct
• Exa AI to search Stack Overflow, GitHub Issues, and developer forums for known gotchas and workarounds

This phase catches the problems that ruin tutorials: the deprecated connectOptions parameter that was renamed in v2.4, the security advisory on the auth middleware, the undocumented rate limit that causes silent failures.

Phase 3 — Tutorial with SEO Optimization

The third agent writes the final tutorial. Before writing, it uses two MCP servers for keyword research:

• SEMrush for keyword difficulty and search volume on the target integration query
• SerpAPI for competitor analysis and People Also Ask questions

The result is a ~3,000-word tutorial with step-by-step instructions, complete code blocks with syntax highlighting, exact version requirements, a troubleshooting section covering the top 5 errors, and a FAQ targeting developer search queries.

---

Why MCP Servers Are the Right Tool for Developer Agents

Developer tooling is fragmented across dozens of platforms — GitHub, GitLab, npm, PyPI, Docker Hub, Snyk, Sentry, SonarCloud, and more. Building custom integrations for each platform is exactly the kind of tedious work that slows down agent development.

The Vinkius AI Gateway provides a managed registry of over 2,600 production-ready MCP servers that cover the entire developer ecosystem. Each server is an authenticated SSE endpoint:

https://edge.vinkius.com/<your_token>/mcp

This project uses 11 of those 2,600+ servers. The architecture is extensible — adding a new data source requires a single configuration change. Consider what becomes possible:

• Dependency audit agents combining Snyk, npm audit, and GitHub Advisory Database in a single crew
• Migration assistants that read the old framework's source via GitHub and the new framework's docs via Firecrawl
• API documentation generators that analyze OpenAPI specs from SwaggerHub and generate client SDKs
• CI/CD pipeline builders pulling deployment patterns from Docker Hub, Vercel, and GitHub Actions
• Code review bots combining SonarCloud quality metrics with Sentry error patterns and Snyk vulnerability data

The Vinkius AI Gateway already has MCP servers for Stripe, Supabase, Vercel, Notion, Linear, Jira, MongoDB, Redis, Datadog, and hundreds more. The limiting factor is no longer integration effort — it is deciding what to build.

---

MCP Servers Used in This Project

| MCP Server | Agent | Data Provided | |---|---|---| | github-mcp | Senior Engineer | Repository code, releases, issues, examples | | gitlab-mcp | Senior Engineer | GitLab repositories, CI/CD configs | | docker-hub-mcp | Senior Engineer | Docker images, tags, deployment configs | | firecrawl-mcp | Senior Engineer | Official documentation, API references | | snyk-mcp | Code Reviewer | Dependency vulnerability scanning | | sonarcloud-mcp | Code Reviewer | Code quality, best practice patterns | | sentry-mcp | Code Reviewer | Common runtime errors from other devs | | swaggerhub-mcp | Code Reviewer | API specification validation | | exa-ai-mcp | Code Reviewer | Stack Overflow, dev forums, blog posts | | semrush-mcp | Technical Writer | Developer keyword research | | serpapi-mcp | Technical Writer | SERP features, competitor tutorials |

All 11 servers are hosted on the Vinkius AI Gateway. Browse the full catalog of 2,600+ production-ready MCP servers at vinkius.com/en/categories.

---

Getting Started

Prerequisites

• Python 3.11+
• A Google Gemini API key (free tier is sufficient)
• MCP server endpoints from the Vinkius AI Gateway

Installation

git clone https://github.com/vinkius-labs/crewai-mcp-dev-tutorial-agents.git
cd crewai-mcp-dev-tutorial-agents

python -m venv .venv
source .venv/bin/activate    # Linux/macOS
# .venv\Scripts\activate     # Windows

pip install -e .

Configuration

cp .env.example .env

Open .env and configure:

1. Your Gemini API key from Google AI Studio
2. Your Vinkius MCP URLs — deploy the MCP servers you need from the Vinkius AI Gateway marketplace, then copy each server's SSE endpoint URL

Usage

# Validate your environment
dev-tutorial validate

# Generate a tutorial
dev-tutorial generate "Stripe" "Supabase" --framework "Next.js 15"

# The tutorial is saved to output/
cat output/stripe-supabase-next-js-15-integration-guide.md

Example Commands

# Stripe + Supabase in Next.js
dev-tutorial generate "Stripe" "Supabase" --framework "Next.js 15"

# Redis + FastAPI with Docker
dev-tutorial generate "Redis" "FastAPI" --framework "Docker"

# Sentry + Django
dev-tutorial generate "Sentry" "Django"

# Auth0 + Express.js
dev-tutorial generate "Auth0" "Express.js" --framework "Node.js"

---

Generated Tutorial Structure

| Section | Purpose | |---|---| | TL;DR | What you will build, in one paragraph | | Prerequisites | Exact versions, accounts, API keys needed | | Project Setup | Terminal commands to scaffold from zero | | Step-by-Step Integration | Complete code blocks with filenames | | Testing | Verification commands for each step | | Error Handling | Production-ready patterns, not demo code | | Deployment Considerations | Env variables, secrets, Docker setup | | Troubleshooting | Top 5 errors with solutions | | FAQ | 5 developer questions targeting search | | Version Reference | Exact package versions tested |

---

Technical Details

• Framework: CrewAI with Flows and @CrewBase decorators
• LLM: Google Gemini 2.0 Flash (free tier, ~15 RPM)
• State Management: Pydantic models for type-safe data flow between agents
• MCP Integration: Native CrewAI mcps= field with SSE transport to Vinkius AI Gateway
• CLI: Typer with Rich console output
• Rate Limiting: max_rpm=10 per agent to stay within Gemini free tier limits

---

FAQ

What is MCP?

The Model Context Protocol is an open standard for connecting AI systems to external tools and data sources. See modelcontextprotocol.io.

Does the engineer agent actually read source code?

Yes. Through the GitHub and GitLab MCP servers, the agent reads file contents, browses directory structures, checks the latest releases, and analyzes real code. Through Firecrawl, it scrapes the official documentation pages. It does not rely on training data — it reads the current state of the code.

Can I use a different LLM?

Yes. CrewAI supports OpenAI, Anthropic, Mistral, and any LiteLLM-compatible model. Change the LLM configuration in crew.py.

How do I add more MCP servers?

Add the server to config/mcp_servers.yaml, set the URL in .env, and it becomes available. The Vinkius AI Gateway offers 2,600+ MCP servers — explore the full catalog.

---

Contributing

We welcome contributions. Please read the Contributing Guide before submitting a pull request.

---

License

MIT — see LICENSE.

---

Built by Vinkius Labs with CrewAI and the Vinkius AI Gateway.