Описание
MCP Servers
README
A comprehensive collection of Model Context Protocol (MCP) servers for Claude Code, Gemini, Copilot, providing integrations with MySQL, Jira, GitHub, CloudWatch, Azure AD, and more.
Setup
git clone https://github.com/teo-lin/multi-llm-mcps cd mcp-servers npm run setup
Edit the .env files in each MCP directory with your credentials
Manually register a single mcp server
## ADD GitHub 0.2k tokens
claude mcp add github --scope user -- npx --yes @teolin/mcp-github
# REMOVE
claude mcp remove github --scope user
Or... Install official ones and see your tokens fly out the window. Context window.
## ADD GitHub 5.1k tokens
claude mcp add github --scope user -- npx --yes @modelcontextprotocol/server-github
## REMOVE
claude mcp remove github --scope user
## ADD Playwright 3.6k tokens
claude mcp add playwright --scope user -- npx --yes @playwright/mcp@latest
## REMOVE
claude mcp remove playwright --scope user
Available MCP Servers
1. MySQL (mysql)
Connect to MySQL databases, execute queries, and manage schemas.
Tools:
query- Execute SQL queries with parameterized placeholderslist_databases- List all databaseslist_tables- List tables in a databasedescribe_table- Show table structure
Configuration: MySQL/.env
2. Jira (jira)
Comprehensive Jira integration with advanced querying capabilities.
Tools:
get_ticket_details- Get ticket informationsearch_tickets_jql- Search using JQLget_board_issues- Get sprint board issuesget_ptls_board_bugs- Get PTLS bugs by teamget_absences_bugs- Get Absences team bugsget_team_names- List team names
Configuration: Jira/.env
3. GitHub (github)
GitHub operations using GitHub CLI.
Tools:
- PR information and diffs
- Repository context extraction
- Multiple PR identifier formats support
Prerequisites: GitHub CLI (gh) authenticated
Configuration: GitHub/.env
4. CodeReview (code-review)
Automated PR code review with Jira integration.
Tools:
codereview- Automated PR analysis with Jira context
Features:
- Fetches PR diffs and metadata
- Retrieves Jira requirements
- Runs tests and linting
- Generates AI-powered review
Prerequisites: GitHub CLI + Atlassian CLI
Configuration: CodeReview/.env
5. Atlassian (atlassian)
Atlassian Jira operations via Atlassian CLI.
Tools:
jira_ticket_info- Get ticket detailsjira_extract_ticket_from_text- Extract ticket keysjira_open_ticket- Open in browserjira_auth_status- Check authentication
Prerequisites: Atlassian CLI (acli) authenticated
Configuration: Atlassian/.env
6. CloudWatch (cloudwatch)
Query AWS CloudWatch Logs using Log Insights.
Tools:
query_logs- Execute CloudWatch Logs Insights querieslist_log_groups- Browse log groupsget_recent_logs- Quick access to recent entries
Prerequisites: AWS credentials configured
Configuration: CloudWatch/.env
7. AzureAD (azuread)
Azure Active Directory authentication with OAuth 2.0.
Tools:
authenticate- Device code flow authenticationget_access_token- Get current tokencheck_auth_status- Check authenticationclear_token_cache- Force re-authenticationmake_authenticated_request- Authenticated HTTP requests
Configuration: AzureAD/.env
8. Kafdrop (kafdrop)
Kafka cluster inspection via Kafdrop Web UI.
Tools:
list_topics- View all topicsget_topic_details- Topic informationbrowse_messages- Read messageslist_consumer_groups- Monitor consumerslist_brokers- List Kafka brokerssearch_messages- Search for messages
Configuration: Kafdrop/.env
9. MSK (msk)
AWS MSK (Managed Streaming for Kafka) with protobuf message decoding.
Tools:
list_clusters- List MSK clustersget_cluster_details- Get cluster informationlist_topics- List Kafka topicsget_topic_metadata- Get partition/offset infobrowse_messages- Browse messages from partition/offsetsearch_messages- Search across partitionslist_protobuf_types- List available protobuf message types
Features:
- Direct AWS MSK integration via AWS SDK
- KafkaJS for Kafka operations
- Protobuf message decoding using
@doctaridev/io.planer.library.npm.protobuf - Support for IAM, TLS, and plaintext authentication
Prerequisites: AWS credentials configured, network access to MSK cluster
Configuration: MSK/.env
Management Scripts
# Setup everything (first time)
npm run setup
# Register all servers
npm run register
# Unregister all servers
npm run unregister
# Re-point the stable symlink after renaming or moving this repo
./scripts/relink.sh
Path independence
Claude Code stores an absolute command path per stdio server, so registrations would
break every time this folder is renamed or moved. To avoid that, servers are registered
under a fixed path — ~/.mcp/mcps/<Server>/start-mcp.sh — where ~/.mcp is a symlink to
this repo. After a rename or move, run ./scripts/relink.sh once; no re-registration and
no config edits needed. register-all.sh calls it automatically. Override the link
location with MCP_LINK=/some/path.
Documentation
Each MCP server has its own README with detailed documentation:
- MySQL README
- Jira README
- GitHub README
- CodeReview README
- Atlassian README
- CloudWatch README
- AzureAD README
- Kafdrop README
- MSK README
Requirements
- Node.js: 25.2.1 (automatically installed by setup script)
- Claude Code: Latest version
- GitHub CLI: For GitHub and CodeReview MCPs (
brew install gh) - Atlassian CLI: For Atlassian and CodeReview MCPs
- AWS CLI: For CloudWatch MCP (credentials configured)
Security
- Never commit
.envfiles to version control - All
.envfiles are in.gitignore - API tokens and credentials are stored locally only
- Each MCP runs in isolated process
Introduction to MCP Servers
What in the Devil's name are those...?
Think of MCP servers as smart-ish middleware between the agent's tools and the agent. An agent is basically a glorified chat-bot with tools. You can create your own coding agent starting from Codellama or DeepSeek by adding a minimum of 4 tools: file reader/editor, shell client, web-clinet, debugger. Think of these as supplementary abilities, besides the chat-bot's skill of diplomatically serving you bullshit.
MCP stands for Model-Context Protocoll, a fancy-ass name for a framework that lets AI agents handle structured inputs (basically JSON), manage context across tasks, and produce more consistent, meaningful outputs.
Now, since you cannot modify the built-in tools that Claude Code or Gemini CLI ship with, you can extend these with your own: the mighty MCP servers.
Think of these as functions you can call. They're like little Lambdas or Google Functions, running locally. There's a plethora of them freely available, which may or may not come with massive security risks. Good luck with those. You can create them with Node, Python, whatever you fancy. You can ask your AI agent to build them for you. It's basically just a one-file project.
Setup: Create a small server app exposing one or more functions, like code_review(pull_request_id), then register the tool so that your agent can "see" them (usually with mcp add my-code_review-server --scope user /path/to/server). Verify with claude mcp list to ensure it shows " Connected".
Run with - every service differs: mcp__code_review_review 1234 (Claude) or cursor.agent.codeReview(pull_request_id) (you guessed it, Cursor) etc. You can also @-mention them to toggle on/off, or use /mcp to see all available tools from connected servers.
Pros: Scoped, accepts params, can authenticate and integrate with external services, very flexible
Cons: more complex, needs a repo to share with the team.
Slow down...
WHAT
| Aspect | Knowledge Base | Driver / Runtime | Workflow | Tool (MCP Servers) | AI Coding Agent |
|---|---|---|---|---|---|
| What | Model (LLM) | API + Query Tool | Shortcuts: Skill/Command | Background dormant process | UI/CLI: AI orchestrator |
| Why | Store knowledge | Access information | Automate operations | External service as tool | Analyze, plan, do, test |
| How | Tokenize & Predict next | Run models, serve API | /cmd syntax calls MCPs |
Listen → Execute → Retry | Select tools → iterate |
| Example | Qwen4, DeepSeek | Ollama, LM Studio | /commit git flow |
mcp-github, mcp-mysql |
Cloud/Local, CLI/IDE/UI |
- AI Agents by Category
| Interface | Cloud (paid/proprietary) | Local |
|---|---|---|
| CLI | Claude Code, Gemini-CLI, Codex | Goose, PicoClaw, OpenClaw |
| IDE / UI | Cursor, Windsurf, Antigravity, Cline | LM Studio, Comfy UI, Auto1111 |
| IDE Extensions | Q, Rovo, MS Copilot, GH Copilot, CodeGPT | Continue |
| Browser | Bolt, Lovable, Vercel v0, Replit | HuggingFace Spaces, Codespaces |
WHY
Anthropic's baby, widely adopted. It's a standardized protocol that connects AI models and coding assistants to external tools and services in a structured, cost-friendly way.
Because you don't wanna:
- Copy-paste credentials and secrets into chat willy-nilly.
- Manually describe API responses a zillion times per day
- Explain your coding slave what and how to use every single time
You want:
- Secure, sandboxed access to tools
- Structured tool descriptions
- Automatic tool selection by AI based on task
- Scalable across multiple AI platforms / agents
HOW
Core Pattern: Tools That Return Minimal, Structured Data
An MCP Server wraps external services (GitHub, databases, APIs) as tools, returning Minimal, Structured Data Typical Structure: ├── tools.js # Define what tools do (name, description, input schema) ├── handlers.js # Implement tool logic (call API/service, parse output) ├── index.js # Wire it up (Server → ListTools → CallTools) └── .env # Keep your passwords and tokens safe (use .gitignore as well)
Core Pattern: Context Window Optimization
| Principle | Goal | How |
|---|---|---|
| Precise Descriptions | AI understands purpose | "Get PR details" not verbose explanation |
| Structured I/O | Prevent errors & min tokens | {title, branch} not raw API response with metadata/headers/status |
| Minimal Result Sets | Keep context lean | LIMIT 100 not all 10k rows |
| Clear Error Messages | Fast failure recovery | {error: "reason", suggestion: "fix"} |
Real Result: GitHub PR queries drop from 2000-10k tokens → 300-800 tokens (75%+ savings).
Resources
- MCP Specification: https://modelcontextprotocol.io/
- Official MCP Servers: https://github.com/modelcontextprotocol/servers
- NPM Packages: https://www.npmjs.com/search?q=%40teolin
Установка Multi Llm Mcps
У этого сервера нет опубликованного пакета — он собирается из исходников. Открой репозиторий и следуй инструкции в README.
▸ github.com/teo-lin/multi-llm-mcpsFAQ
Multi Llm Mcps MCP бесплатный?
Да, Multi Llm Mcps MCP бесплатный — установка в пару кликов через Unyly без оплаты.
Нужен ли API-ключ для Multi Llm Mcps?
Нет, Multi Llm Mcps работает без API-ключей и переменных окружения.
Multi Llm Mcps — hosted или self-hosted?
Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.
Как установить Multi Llm Mcps в Claude Desktop, Claude Code или Cursor?
Открой Multi Llm Mcps на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.
Похожие MCP
Fetch
Web content fetching and conversion for efficient LLM usage.
AWS KB Retrieval
Retrieval from AWS Knowledge Base using Bedrock Agent Runtime.
автор: modelcontextprotocolSpring AI MCP Server
Provides auto-configuration for setting up an MCP server in Spring Boot applications.
llm-analysis-assistant
A very streamlined mcp client that supports calling and monitoring stdio/sse/streamableHttp, and can also view request responses through the /logs page. It also
автор: xuzexin-hzCompare Multi Llm Mcps with
Не уверен что выбрать?
Найди свой стек за 60 секунд
Автор?
Embed-бейдж для README
Похожее
Все в категории ai
