Caboose MCP
A comprehensive personal AI toolserver that exposes 105 tools across 25 groups—including Slack, Docker, and calendar management—to Claude and VS Code. It features a Go-based MCP server integrated with a custom VS Code extension and n8n workflow automation for streamlined personal productivity.
README
fafb
Personal AI toolserver — 129 MCP tools exposed to Claude, VS Code, and chat bots via a Go server hosted on AWS ECS, with Discord OAuth, admin approval workflows, and automatic tool synchronization.
Something selfishly for me but hopefully useful for others.
⚡ NEW: Ecosystem Manager (Phases 1-6) — Discord OAuth, auto-deployment, tool creation UI, admin approval workflows, and 129 tools extracted and synchronized.
Live endpoint: https://mcp.chrismarasco.io/mcp · Web UI: https://mcp.chrismarasco.io/ui/
[!WARNING] Experimental software — use at your own risk. This project is under active development and has not been fully tested. Tools may behave unexpectedly, modify data, or fail without warning. See the Disclaimer for full terms.
Architecture
graph TD
Claude["Claude Code (stdio)"] -->|all 118 tools| Binary["fafb (Pi)"]
VSCode["VS Code / Bruno"] -->|HTTPS| ALB["ALB · mcp.chrismarasco.io"]
ALB --> Serve["ECS · --serve-hosted\n68 hosted tools"]
Slack["Slack"] --> Bots["ECS · --bots\nSlack + Discord gateway"]
Discord["Discord"] --> Bots
Bots -->|Claude Haiku| Agent["Bot agent loop"]
Binary -->|LAN MQTT| Bambu["Bambu A1 Printer"]
Binary -->|socket| DockerD["Docker daemon"]
Serve & Bots --> SM["AWS Secrets Manager\nfafb/env"]
Tool Tiers
Tools are split so the cloud server only exposes what's safe remotely.
| Tier | Flag | Count | What's included |
|---|---|---|---|
| Hosted | --serve-hosted |
68 | Calendar, Slack, Discord, GitHub, Notes, Focus, Learning, Sources, CloudSync, Audit, Auth, Health, Secrets, DB, Env, Mermaid, Greptile, Sandbox, Persona, Jokes, Setup, Claude Files, Self-Improve, Repo Management (create, test, approve, deploy), Gamma (presentations) |
| Local | --serve-local |
25 | Docker, execute_command, Bambu, Blender, Chezmoi, Toolsmith, Agency |
| Common | (any mode) | 3 | Jokes, Claude Files, Setup |
| Combined | --serve / stdio |
129 | Everything (common + hosted + local + new tools) |
Full reference: docs/tools.md · ECOSYSTEM.md
⚡ Ecosystem Manager (NEW)
Phases 1-6 Complete: Secure, integrated toolkit for managing 129 MCP tools with Discord OAuth, admin approval workflows, and automatic tool synchronization.
Key Features
✅ Phase 1: Discord OAuth2 PKCE flow for user authentication ✅ Phase 2: Auto-extract 129 tools + auto-sync UI on push to main ✅ Phase 3: Tool creation, testing, and admin approval workflow ✅ Phase 4: Auto-generate Gamma presentation decks ✅ Phase 5: GitHub PR-based tool submission with CI tests ✅ Phase 6: Admin UI forms + parameter validation
Create a Tool
Via Web UI:
1. Go to https://ui.mcp.chrismarasco.io/admin/create-tool
2. Fill in tool details (name, description, parameters)
3. Click "Create Tool" → saved to ~/.claude/pending-tools/
4. Test via /admin/pending-tools
5. Admin approves → commits to main → auto-deploys
Via MCP:
fafb repo_create_tool --name "my_tool" --description "Does X" --tier "hosted"
fafb repo_test_tool --tool_name "my_tool" --test_input_json '{}'
fafb repo_approve_tool --tool_name "my_tool"
See ECOSYSTEM.md for complete workflow and API reference.
Quick Start
Connect to the live server
Claude Code CLI:
# HTTP — works in any project
claude mcp add --transport http MCP_CABOOSE https://mcp.chrismarasco.io/mcp
# User scope — available across all projects
claude mcp add --transport http --scope user MCP_CABOOSE https://mcp.chrismarasco.io/mcp
VS Code — add to .vscode/mcp.json:
{
"servers": {
"fafb": {
"type": "http",
"url": "https://mcp.chrismarasco.io/mcp"
}
}
}
Bruno — open the bruno/ folder as a collection.
Run locally (Claude Code on Pi)
cd packages/server && go build -o fafb .
# Interactive setup wizard — writes .env
./fafb --setup
# Terminal UI dashboard
./fafb --tui
./fafb # stdio — all tools (Claude Code)
./fafb --serve # HTTP :8080 — all tools
./fafb --serve-hosted # HTTP :8080 — hosted tools only
./fafb --serve-local # HTTP :8080 — local tools only
./fafb --bots # Slack + Discord bots (blocks)
.mcp.json for Claude Code:
{
"mcpServers": {
"caboose": { "command": "/path/to/packages/server/fafb", "type": "stdio" }
}
}
Or via CLI:
claude mcp add MCP_CABOOSE -- /path/to/fafb
Run with Docker Compose (server + n8n)
cp .env.example .env # fill in secrets
docker compose -f docker/docker-compose.yml up -d
| Service | URL |
|---|---|
| MCP server | http://localhost:8080/mcp |
| n8n | http://localhost:5678 |
Full setup instructions for both local and hosted deploys: docs/setup.md
CI / CD
| Workflow | Trigger | Action |
|---|---|---|
ci.yml |
PR / push to main | Lint, test, build (amd64+arm64), e2e, UI build + changelog generation |
deploy-infra.yml |
Push to terraform/aws/** or manual |
tofu apply + sync secrets to AWS Secrets Manager |
deploy-bots.yml |
Push to packages/server/** on main |
Build image → ECR → redeploy fafb-bots |
deploy-app.yml |
Push to packages/server/** or docker/Dockerfile on main |
Build image → ECR → redeploy fafb-serve → index Greptile |
release.yml |
Push to main or v*.*.* tag |
Build linux/amd64 + linux/arm64 → GitHub Release (auto date-tag on merge) |
Releases
Releases are created automatically on every merge to main with a date-based tag (vYYYY.MM.DD.N). For a versioned release, push a v*.*.* tag:
git tag v1.2.0
git push origin v1.2.0
Infrastructure
AWS resources managed by OpenTofu in terraform/aws/:
- ECS Fargate —
fafb-bots(--bots) +fafb-serve(--serve-hosted) - ALB — HTTPS 443, HTTP→HTTPS redirect,
mcp.chrismarasco.io - ACM — TLS cert, DNS-validated via Route53
- ECR — Docker image registry (lifecycle: keep last 5)
- Secrets Manager —
fafb/env— secrets injected into ECS tasks at startup - S3 — encrypted config sync bucket
- CloudWatch Logs — 30-day retention per service
cd terraform/aws
cp terraform.tfvars.example terraform.tfvars
tofu init && tofu plan && tofu apply
Adding a Tool
tool_scaffold # generate tools/mytool.go skeleton
tool_write # write the file
tool_rebuild # go build
Or edit packages/server/tools/ directly — one .go file per feature group.
Repository Layout
packages/server/ Go MCP server (117 tools)
tools/ One .go file per feature group
config/config.go All env vars → Config struct
main.go Flags, server builders, HTTP mux, bot runner
packages/vscode-extension/ VS Code extension
bruno/ Bruno collection (120 requests, 24 categories)
docker/Dockerfile Multi-stage: Go build → alpine runtime
terraform/aws/ OpenTofu — ECS, ALB, ACM, ECR, S3, Secrets Manager
.github/workflows/ ci, deploy-infra, deploy-bots, deploy-app, release
docs/
tools.md Full tool reference
setup.md Local + hosted deployment guide, JWT RBAC, AWS costs
Disclaimer
fafb is experimental software provided "as is", without warranty of any kind, express or implied.
By using this software you acknowledge and agree that:
- This project is under active development and has not been fully tested in all environments or configurations.
- Tools may execute shell commands, modify files, query external APIs, send messages, or interact with cloud infrastructure. Outcomes are not guaranteed to be correct or safe.
- No warranty is provided — express, implied, statutory, or otherwise — including but not limited to warranties of merchantability, fitness for a particular purpose, or non-infringement.
- The author(s) shall not be liable for any direct, indirect, incidental, special, consequential, or exemplary damages arising from the use or inability to use this software, even if advised of the possibility of such damages.
- Do not use in production systems, with sensitive or regulated data, or in any context where failures could cause harm, until you have reviewed and validated the relevant tools for your use case.
This disclaimer is in addition to and does not limit the terms of the MIT License.
The FAFB_ENV=stable flag may be set to suppress experimental warnings once a deployment has been reviewed and accepted for a given use case — this does not alter or waive any of the above terms.
License
MIT — Copyright (c) 2025 Chris Marasco
Credits
Built with love, chaos, and coffee by:
| Contributor | Role |
|---|---|
| Chris Marasco (@cxm6467) | Architect, product owner, primary developer |
| Claude (Anthropic) | AI pair programmer and code generation |
| Claude Code | Agentic CLI — implementation, refactoring, debugging |
| GitHub Copilot | Inline completion and suggestions |
Thanks bear for the name idea!
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。