Stateful task engine + AI-native tool registry + execution. Part of the Agent Tool Platform.
AgentPilot is an MCP (Model Context Protocol) server that gives AI agents persistent, auditable task infrastructure: task trees in SQLite, deterministic state machine, tool registry with semantic search integration, and autonomous execution capability.
| Module | What it does | Status |
|---|---|---|
| Agent Tool Intel | Quality scoring + search + trust for MCP tools | ✅ |
| AgentPilot ← you are here | Task orchestration + tool registry + execution | ✅ |
| AutoMine | Automated tool discovery from content | 🔧 |
All three share a unified Canonical ID system: tool:{source}:{namespace}/{name}@version
git clone https://github.com/agent-tool-intel/agent-pilot.git
cd agent-pilot
npm install
npm run build
npm startRuns on stdio. Auto-creates SQLite database on first launch.
Agent request → State Machine → Task Tree → Tool Registry → Execute → Feedback
│ │ │
▼ ▼ ▼
SQLite DB FTS Search Agent Tool Intel API
(sql.js) (LIKE) (quality scores)
- State Machine — Deterministic task state transitions (pending → in-progress → done/failed)
- Task Trees — Parent/child task hierarchy with dependency tracking
- Tool Registry — Persistent tool store with canonical IDs and Agent Tool Intel integration
- Model Router — Zero-LLM rule engine for task-to-model routing
- Audit Log — Full state change tracking with timestamps
- 30 MCP tools for task orchestration
- SQLite-backed persistence (sql.js — pure JS, no native deps)
- FTS-based tool search
- Automated task status transitions
- Snapshot/restore for task trees
- Archival system for completed tasks
- Model configuration per task category
- Batch task operations
AgentPilot integrates with Agent Tool Intel for:
- Tool quality scoring
- Semantic tool search
- Execution analytics tracking
MIT