MCPolly
About
Agent-native observability for AI agents. MCP server with web dashboard, alerting, and semantic search. Built with Rust, Axum, SQLite, and HTMX.
Details
- Author
- MCPolly
- GitHub stars
- 2
- Downloads
- 130
- Categories
- Developer Tools
Jump to
- Agent registration and real‑time status tracking (starting, running, error, completed, etc.)
- Real‑time activity feed and global error feed across all agents
- Configurable webhook alerts (Discord, Slack, generic) with retry logic
- Dark‑mode web dashboard with live updates via HTMX polling
- Vector semantic search over indexed documents using local Ollama embeddings
- Agent spawning with contextual knowledge from indexed content
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
MCPollyCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
Install via the install script or build from source, then start the server with PORT=3000 RUST_LOG=mcpolly=info ./mcpolly. On first run, a setup wizard in the browser provides a default API key. Configure your AI agent platform (e.g., Cursor, Claude Code) to connect to the server’s /mcp endpoint using HTTP Bearer auth or the optional stdio bridge binary. Agents can then use MCP tools like register_agent, post_status, and post_error to report activity.
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"mcpolly": {
"mcpolly": {
"url": "http://localhost:3000/mcp",
"headers": {
"Authorization": "Bearer your-api-key-here"
}
}
}
}
}McpServers
{
"mcpolly": {
"url": "http://localhost:3000/mcp",
"headers": {
"Authorization": "Bearer your-api-key-here"
}
}
}MCPolly
Agent-native status, observability, and knowledge platform for AI agents. MCPolly is an MCP server that lets AI agents report their progress, errors, and state in real time — and gives humans a unified web dashboard to monitor everything. Agents can also index and semantically search product documents to stay aligned.
Built with Rust, Axum, SQLite, and HTMX. Designed to self-host on minimal hardware.
Architecture
┌─────────────────┐ MCP Streamable HTTP ┌──────────────────┐
│ AI Agent │◄───────────────────────►│ MCPolly Server │
│ (Cursor, etc.) │ JSON-RPC / SSE │ (Axum + SQLite) │
└─────────────────┘ Bearer auth └──────┬───────────┘
│
┌────────▼────────┐
│ Web Dashboard │
│ (HTMX UI) │
└─────────────────┘
│
┌────────▼────────┐
│ Ollama (local) │
│ all-MiniLM │
└─────────────────┘
MCPolly exposes a /mcp endpoint that speaks the MCP Streamable HTTP protocol (JSON-RPC over HTTP with SSE streaming). Agent platforms connect directly — no subprocess binary needed.
- MCPolly Server — The Axum HTTP server that stores agent data in SQLite, serves the web UI, evaluates alert rules, manages vector embeddings, and hosts the MCP endpoint.
- Ollama — Local LLM server used for generating vector embeddings with the all-MiniLM model (384 dimensions).
- mcpolly_mcp (optional) — A lightweight stdio binary for platforms that don't support HTTP MCP transport. It bridges stdio to the MCPolly HTTP API.
Features
Agent Observability
- Agent registration and status tracking (starting, running, warning, error, completed, offline, paused, errored, stopping, stopped) - Real-time activity feed with color-coded entries - Global error feed across all agents - Webhook alerts (Discord, Slack, generic) with retry logic - Configurable alerts for any status change — error, completed, running, starting, warning, paused, stopped, offline, or a catch-all "any status" rule - Silent agent detection (background checker)Web Dashboard
- Dark mode with system preference detection - First-run setup wizard with auto-login and MCP configuration snippets - Instance reset from the login page (revokes all keys, generates a new one, re-enters setup wizard) - Health strip summary (total/running/errored/offline agent counts) - Status filter pills for quick agent filtering - Agent detail page with tabbed content (Activity, Errors, Info) - Global command bar (Cmd+K) searching agents, errors, and knowledge - Alert rules management with notification history - Knowledge page with semantic search UI - API key management - Settings page with server info and session management - 10-second HTMX polling for live updatesKnowledge Layer (Vector Embeddings)
- Index PRD, design, and custom documents as vector embeddings - Semantic search across all indexed content via MCP tools and web UI - Spawn product manager and product designer agents with relevant context - Powered by sqlite-vec (in-process) and Ollama (local, no cloud API keys needed)Installation
Option 1: Install Script (recommended)
Detects your OS and architecture, downloads the correct binaries from GitHub Releases:
curl -fsSL https://raw.githubusercontent.com/MCPolly/mcpolly/main/install.sh | bash
Customize with environment variables:
```bash
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
