Slash / MCP Server – Technical Product Requirements Document (PRD)
🚀 Installation & Quick-Start (Local Dev)
Prerequisites
- Node.js ≥ 20 and npm ≥ 9 (or pnpm/yarn)
- Git (to clone the repo) – or download the full source zip including
backend/, src/, services/, providers/ folders.
1 – Clone & install
# grab the code
git clone https://github.com/your-org/mcp-server.git
cd mcp-server/project
# install front-end + workspace deps (root package.json)
npm install
# install backend-specific deps npm install --prefix backend
Copy .env.example → .env and fill in at least the values marked (required):
Key vars for a first run:
# ─── Front-end (Vite) ───────────
VITE_SUPABASE_URL=stub
VITE_SUPABASE_ANON_KEY=stub
# ─── Back-end (Express) ────────
OPENAI_API_KEY=your-key-here # optional if you just explore the UI
CORS_ORIGINS=http://localhost:5173 # allow Vite dev server
SESSION_SECRET=dev-secret
ENCRYPTION_KEY=$(openssl rand -hex 32) # 64-char hex
3 – Run dev servers (two tabs)
# tab 1 – front-end (Vite + React)
npm run dev # or pnpm run dev
# tab 2 – back-end (Express)
npm run dev:backend # = cd backend && npm start
Visit http://localhost:5173 – the admin UI talks to the backend on port 3001.
Tip: npm run dev:all starts both processes concurrently.
4 – Build for production
npm run build # front-end → dist/
# serve dist/ with any static server & keep backend running with pm2/systemd
🌟 Why use MCP Messenger Server?
- Unified AI layer – OpenAI, Anthropic, Gemini & more via one
/api/chat contract. - Plug-in providers – GitHub, Gmail, Slack, Jira, Notion, Drive… drop-in JS modules under
backend/providers/. - Secure credential vault – AES-256-GCM encryption + SQLite/Supabase DAO.
- Chained workflows – Parse natural language into multi-step provider commands.
- Batteries included UI – React admin panel for keys, OAuth connects, request history.
- Dev-friendly – Vite hot-reload, CORS allow-list via
CORS_ORIGINS.
1. Vision
A unified backend service for the Model Context Protocol (MCP), enabling API-driven, multi-provider AI and automation command execution, workflow chaining, and seamless integration with services like GitHub and Google. MCP Server acts as the core engine for orchestrating commands and workflows across apps and protocols.
2. Tech Stack
Backend:
- Node.js (Express, ES modules)
- AI Providers: OpenAI, Anthropic, Google Generative AI (Gemini)
- Integrations: Google (OAuth), GitHub (token-based), Slack, Jira, Notion, Loveable, Bolt, 21st.dev
- Session management: express-session
- Database: SQLite (per-user integration tokens)
- Security: CORS, dotenv, multer
Frontend (Admin/Optional):
- React 18 (TypeScript), Vite, Tailwind CSS (dark mode)
- Custom UI components, lucide-react icons
- State management with custom React hooks
3. Core Features
API/Backend
- API endpoints:
/api/chat, /api/workflow, /api/github, /api/google, /api/health - Modular provider/command registry for extensibility
- Multi-step workflow execution and error handling
- File upload and summarization
- Smart intent parsing: routes commands to the correct provider or integration
- Chained command parsing: users/apps can enter multi-step instructions in one message
- Session management and per-user integration tokens
- Secure OAuth flows for Google
- Token-based authentication for GitHub
- Real-time health check endpoint for server status
(Optional) Admin Frontend
- Settings modal for API key entry, provider connection, command reference
- Request history and command palette
- Provider status indicators (connected/disconnected, last used, request count)
- Modern, responsive UI with dark/light mode
4. User Workflows
- Multi-provider AI chat: Apps or users can send chat, summarize, or code commands to OpenAI, Anthropic, or Gemini using their own API keys.
- Command registry: Supports chat, summarize, code search, repo summary, explain, translate, and more.
- Integrations: Google Drive, Gmail, and GitHub via OAuth or token-based credentials.
- Smart intent parsing: Natural language commands are routed to the correct provider or integration.
- Chained commands: Apps or users can enter multi-step instructions in one message (e.g., "Summarize this PDF and translate to Spanish").
- Workflow builder: (Optional) Visual builder for multi-step, multi-provider workflows.
- File upload: Summarize or process uploaded files.
- Session-based authentication: Secure, per-user integration tokens.
5. Security & Best Practices
- TypeScript everywhere
- Tailwind CSS for utility-first, responsive design (frontend)
- OAuth for secure integrations
- Environment variables for secrets/config
- API error handling and user feedback
- CORS restricted to local dev ports
- Per-user token storage in SQLite
6. Roadmap & Next Steps
- Visual workflow builder (drag-and-drop, optional)
- More integrations (Notion, Slack, Jira, etc.)
- Custom command macros and user-defined workflows
- Scheduling and automation
- Contextual memory and chaining across sessions
- Harden OAuth flows, encrypt API keys, audit CORS
- Add unit and integration tests
- Finalise Bolt & 21st.dev integrations; add UI for new providers
7. Example Use Cases
- "Summarize this PDF and then translate it to Spanish."
- "Fetch the latest GitHub issues, summarize them, and email the summary to my team."
- "Extract text from an image, then run sentiment analysis."
Slash / MCP Server
Product Requirements Document (PRD)
Core Features:
- API-driven, multi-provider AI chat (OpenAI, Anthropic, Gemini) with per-user API keys.
- Command registry: supports chat, summarize, code search, repo summary, explain, translate, and more.
- Google Drive, Gmail, and GitHub integrations with OAuth or token-based credentials.
- Real-time provider connection status via API.
- Smart intent parsing: routes commands to the correct provider or integration.
- Chained command parsing: apps or users can enter multi-step instructions in one message.
- Workflow builder (optional): design and run multi-step, multi-provider workflows.
- File upload and summarization.
- Session-based authentication for integrations.
- Modern, responsive (optional) UI for admin/configuration.
Advanced/Upcoming:
- Visual workflow builder (drag-and-drop, optional).
- More integrations (Notion, Slack, Jira, etc).
- Custom command macros and user-defined workflows.
- Scheduling and automation.
- Contextual memory and chaining across sessions.
Features
- API-driven multi-provider AI chat: OpenAI, Anthropic, Gemini (per-user API keys)
- Command registry: Chat, summarize, code search, repo summary, explain, translate, and more
- Integrations: Google Drive, Gmail, GitHub, Slack, Jira, Notion (OAuth or token-based credentials)
- Smart intent parsing: Natural language commands routed to the right provider/integration
- Chained commands: Enter multi-step instructions in one message
- Workflow builder: (optional) Visual multi-step, multi-provider workflows
- File upload: Summarize or process uploaded files
- Session-based authentication: Secure, per-user integration tokens
- Modern UI (optional): Responsive, dark/light mode, real-time provider status
Chained Commands & API
Overview
Chained commands allow apps or users to combine multiple actions in a single workflow, where the output of one command becomes the input for the next. This enables complex, multi-step tasks to be performed with a single API call or through a visual builder.
User Value
- Automates repetitive or multi-step tasks.
- Saves time and reduces manual effort.
- Enables advanced use cases (e.g., summarize a document, then translate the summary).
Example Use Cases
- "Summarize this PDF and then translate it to Spanish."
- "Fetch the latest GitHub issues, summarize them, and email the summary to my team."
- "Extract text from an image, then run sentiment analysis."
Supported Chains
- Summarize → Translate
- Fetch Issues → Summarize → Email
- Extract Text → Analyze Sentiment
Error Handling
- If a step fails, the user/app is notified with a clear error message.
- The user/app can view which step failed and retry or edit the chain.
Chained Command API/GUI
Key Features:
- API for Chained Commands:
- Apps can POST a workflow (array of steps) to
/api/workflow. - Each step specifies provider, command, prompt, and (optionally) API key.
- (Optional) Command Builder Modal/Panel:
- Users can visually add, remove, and reorder steps in a workflow.
- Each step is represented as a card or block with editable parameters.
- Inline Chaining in Chat:
- Users can type chained commands naturally; the system parses and displays the steps visually above the input.
- Users can click to edit the chain in the builder modal.
- Execution Feedback:
- Progress bar or step tracker shows which step is running, completed, or failed.
- Users can expand/collapse to view intermediate results.
User Flow:
- App or user sends a chained command or opens the builder.
- System parses and displays the steps.
- User/app reviews/edits the steps.
- User/app clicks "Run."
- API/UI shows progress and results for each step.
For questions, contributions, or commercial use, contact automationalien.com
MCP API Reference
List Supported Commands
Returns all commands the MCP server currently exposes. Example response (truncated):
[
{ "id": "chat", "description": "Conversational AI" },
{ "id": "summarize", "description": "Summarize text or documents" },
{ "id": "generate-code", "description": "Generate code snippets" },
{ "id": "explain", "description": "Explain concepts or code" },
{ "id": "translate", "description": "Translate text" },
{ "id": "list-repos", "description": "List GitHub repositories", "providers": ["github"] },
{ "id": "get-file", "description": "Fetch file content from repository","providers": ["github"] },
{ "id": "repo-summary", "description": "Summarize repository structure", "providers": ["github"] },
{ "id": "code-search", "description": "Search code in repositories", "providers": ["github"] },
{ "id": "generate-issue", "description": "Create GitHub issue", "providers": ["github"] },
{ "id": "generate-pr", "description": "Create GitHub pull-request", "providers": ["github"] }
]
List Providers & Their Capabilities
Each provider entry lists the commands it supports:
[
{ "id": "openai", "supportedCommands": ["chat","summarize","generate-code","explain","translate"] },
{ "id": "anthropic", "supportedCommands": ["chat","summarize","generate-code","explain","translate"] },
{ "id": "gemini", "supportedCommands": ["chat","summarize","generate-code","explain","translate"] },
{ "id": "github", "supportedCommands": ["list-repos","get-file","repo-summary","code-search","generate-issue","generate-pr"] },
{ "id": "slack", "supportedCommands": ["send-message","list-channels","get-channel-history"] },
{ "id": "jira", "supportedCommands": ["list-projects","create-issue"] },
{ "id": "notion", "supportedCommands": ["list-databases","query-database"] },
{ "id": "loveable", "supportedCommands": ["summarize","translate"] },
{ "id": "bolt", "supportedCommands": ["summarize","translate"] },
{ "id": "21st.dev", "supportedCommands": ["summarize","translate"] }
]
Use these endpoints to dynamically build command palettes or validate which operations are available without hard-coding them in the front-end.
The project now ships with first-class, open-source connectors that let anyone orchestrate MCP commands and workflows in their favorite no-code tools.
| Platform | Location in Repo | What's Included |
|---|
| n8n | integrations/n8n-nodes-mcp/ | Custom node MCP with Command & Workflow resources and credential type (Base URL + optional API key). Build via npm run build and install inside your n8n instance with npm i n8n-nodes-mcp. |
| Make.com | integrations/makecom/mcp-app.json | Importable JSON definition exposing Execute Command & Execute Workflow actions. |
| Zapier | integrations/zapier-mcp/ | Zapier Platform CLI app with custom auth, ping trigger, and the same two actions. Deploy with zapier push. |
All three connectors share the identical credential model:
- Base URL – root of your MCP server (e.g.
https://api.my-mcp.com) - API Key – optional; include only if your instance enforces key-based auth
Quickstart
# n8n (inside your n8n container / directory)
npm i n8n-nodes-mcp && n8n restart
# Make.com
# 1. Open Make.com ➜ Apps ➜ Create Custom App ➜ Import JSON
# 2. Paste the contents of integrations/makecom/mcp-app.json
# Zapier
npm i -g zapier-platform-cli
cd integrations/zapier-mcp
npm install
zapier login
zapier push # follow prompts to create a private version
See the individual READMEs in each folder for more details.
Claude Desktop Integration – Quick Start
Built-in Zapier Provider (AI Actions)
MCP Server now supports Zapier AI Actions (NLA) natively via the /api/command endpoint.
- In Zapier, open your Zap ➜ AI Actions panel ➜ enable Expose this Zap to AI ➜ Publish the Zap so it's On.
- Generate an AI-Actions API key (it starts with
sk-nla- or legacy sk-ak-). - Call MCP with
provider="zapier" and the key in the body (apiKey) or an Authorization: Bearer … header.
# List AI-enabled Zaps
curl -X POST -H "Content-Type: application/json" \
-d '{"provider":"zapier","command":"list-zaps","apiKey":"<ZAPIER_NLA_KEY>"}' \
http://localhost:3001/api/command
# Trigger a Zap (replace <ID> with one returned from list-zaps)
curl -X POST -H "Content-Type: application/json" \
-d '{"provider":"zapier","command":"trigger zap <ID>","apiKey":"<ZAPIER_NLA_KEY>"}' \
http://localhost:3001/api/command
The server converts both requests into calls to Zapier's NLA endpoints:
list-zaps → GET https://nla.zapier.com/api/v1/ai_zaps/trigger zap <id> → POST https://nla.zapier.com/api/v1/ai_zaps/<id>/execute/
All responses (including errors) are returned as JSON, making them safe to consume from the front-end or other automations.
Jira Integration (OAuth 2.0 3-LO)
The MCP Server now supports Atlassian Jira Cloud commands out-of-the-box.
Supported slash commands
| UI slash | Prompt slug | Description |
|---|
/jira list projects | list-projects | List accessible Jira projects |
/jira list issues | list-issues | Recent issues (default assigned) |
/jira get issue ABC-123 | get-issue ABC-123 | Full issue JSON |
/jira create issue KEY "Summary" | create-issue KEY "Summary" | Create new issue |
/jira update issue ABC-123 … | update-issue ABC-123 … | Update fields (JSON body) |
/jira add comment ABC-123 "…" | add-comment ABC-123 "…" | Append comment & return id |
/jira transition issue ABC-123 Done | transition-issue ABC-123 Done | Move issue to status |
Backend env-vars ( project/backend/.env )
PORT=3001 # keep 3001 for Google + Jira in dev
ATLASSIAN_CLIENT_ID=<app id>
ATLASSIAN_CLIENT_SECRET=<secret> # if confidential mode enabled
JIRA_REDIRECT_URI=http://localhost:3001/api/auth/jira/callback
JIRA_SCOPES=read:jira-work write:jira-work offline_access
CORS_ORIGINS=http://localhost:5174,http://localhost:5173
Atlassian developer-console set-up
- OAuth 2.0 (3-LO) app → Authorisation → Redirect URIs → add the value of
JIRA_REDIRECT_URI. - Copy the Client ID + Secret into
.env. - Grant the scopes shown above (or a subset) and Save.
Supabase row
Update the mcp_servers table so Jira points at the correct port in dev:
update mcp_servers
set apiurl = 'http://localhost:3001'
where id = 'jira';
(Production deployments should point to the public MCP Server URL.)
After these steps "Connect Jira" in the Integrations portal opens the Atlassian consent screen; once accepted the badge flips to Connected and slash commands work via the MCP protocol.