hive-exp
An MCP server enabling AI agents to record, query, and share structured problem-solving experiences with human review and confidence decay.
README
hive-exp
AI Agent Experience Management System — structured, cross-agent, human-reviewed knowledge for your AI tools.
<p align="center"> <strong>English</strong> | <a href="README_CN.md">简体中文</a> </p>
What is hive-exp?
When an AI agent solves a non-trivial problem — a TypeScript compiler error, a broken build, a failing test — the solution disappears the moment the session ends. The next time the same error appears (in a different project, by a different agent, even by the same agent tomorrow), the work starts from zero. hive-exp fixes this by recording experiences: structured JSON objects with a canonical signal → strategy → outcome shape, persisted locally and queryable in milliseconds.
Knowledge does not stay siloed per-agent. Because hive-exp speaks the Model Context Protocol (MCP), Claude Code, Codex, Gemini CLI, Cursor, and Windsurf all read and write the same experience store. An experience recorded by Claude Code while fixing a TypeScript path alias issue is immediately available to Codex the next time it encounters the same signal. The experience store is a shared brain, not a per-agent scratch pad.
Humans stay in the loop. Experiences start as provisional and can only be promoted to the trusted zone by an explicit human action — via the CLI or the dashboard. Confidence decays over time using an exponential half-life model; experiences that accumulate consecutive failures or go 30 days without a single reference are auto-archived. The dashboard gives you a live view of every experience, its current confidence, and its usage statistics.
Quick Start
# Install CLI globally
npm install -g hive-exp
# Initialize for your AI agents (auto-detects installed agents)
hive-exp init --force
# Or run the MCP server directly
npx @hive-exp/mcp
Agent Configuration
hive-exp init --force writes the correct snippet automatically. To configure manually:
Claude Code — ~/.mcp.json
{
"mcpServers": {
"hive-exp": {
"command": "npx",
"args": ["-y", "@hive-exp/mcp"]
}
}
}
Codex — ~/.codex/config.toml
[mcp_servers.hive-exp]
type = "stdio"
command = "npx"
args = ["-y", "@hive-exp/mcp"]
Gemini CLI — ~/.gemini/mcp.json
{
"mcpServers": {
"hive-exp": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@hive-exp/mcp"],
"env": {}
}
}
}
Antigravity — ~/.gemini/antigravity/mcp_config.json
{
"mcpServers": {
"hive-exp": {
"command": "npx",
"args": ["-y", "@hive-exp/mcp"]
}
}
}
Cursor — .cursor/mcp.json
{
"hive-exp": {
"command": "npx",
"args": ["-y", "@hive-exp/mcp"]
}
}
Windsurf — .windsurf/mcp.json
{
"hive-exp": {
"command": "npx",
"args": ["-y", "@hive-exp/mcp"]
}
}
CLI Commands
| Command | Description |
|---|---|
hive-exp init [--force] [--agent <type>] |
Auto-detect AI agents and write MCP configuration |
hive-exp add [--file <path>] [--signals ...] [--strategy <name>] |
Add a new experience record |
hive-exp validate <path> |
Validate an experience JSON file against the schema |
hive-exp sign <path> [--secret <secret>] |
Sign an experience file with HMAC-SHA256 |
hive-exp query [--signal] [--strategy] [--scope] [--limit] [--format] |
Query experiences by signal, strategy, or scope |
hive-exp promote <exp_id> [--confirm] |
Promote an experience to the trusted zone (human confirmation required) |
hive-exp archive <exp_id> [--reason] |
Archive an experience (soft delete) |
hive-exp stats [--type] [--format] |
Show strategy statistics and experience health overview |
hive-exp replay [--from <date>] [--verbose] |
Rebuild SQLite projection from the event log |
hive-exp export [--format] [--min-confidence] [--scope] [--agent] [--promoted-only] [--output] |
Export experiences for RAG or external consumption |
MCP Tools
The MCP server exposes five tools to connected agents:
| Tool | Description |
|---|---|
hive_exp_query |
Search for experiences matching error signals; returns strategies ranked by success rate and confidence |
hive_exp_record |
Record a new experience after successfully solving a non-trivial problem |
hive_exp_outcome |
Record the outcome after applying a strategy from a queried experience |
hive_exp_stats |
Get strategy statistics and experience health overview (overview, strategy_ranking, at_risk) |
hive_exp_promote |
Propose promoting an experience to the trusted zone (sets pending_promotion; actual promotion requires human confirmation) |
Dashboard
# Start the dashboard (requires the CLI to be installed)
hive-exp dashboard
# Or start directly from source
npx tsx apps/dashboard/src/server.ts
The dashboard runs at http://localhost:3333 and shows all experiences, their current confidence scores, usage statistics, and a promotion queue for human review.
Architecture
packages/core/ — Core library (schema, events, signer, sanitizer, consensus, cron)
packages/mcp/ — MCP Server (5 tools, stdio transport, zero external dependencies)
packages/signer-ed25519/ — Optional Ed25519 signer (drop-in replacement for HMAC-SHA256)
apps/cli/ — CLI tool (10 commands)
apps/dashboard/ — Web dashboard (Express + HTML/CSS/JS)
hooks/ — Claude Code PostToolUse hook (signal-detector.py)
Data is stored under ~/.hive-exp/ by default:
~/.hive-exp/
├── experiences/
│ ├── provisional/ — New, unreviewed experiences
│ ├── promoted/ — Human-confirmed trusted experiences
│ ├── archived/ — Auto-archived (zero-ref, low-confidence, consecutive fail)
│ └── superseded/ — Dedup-superseded experiences (backup)
├── events/ — Append-only JSONL event log (yyyy-mm.jsonl)
├── db.sqlite — SQLite projection for fast queries
└── signal-conventions.yaml
How It Compares
| Feature | hive-exp | Mem0 | Letta | Vector DB |
|---|---|---|---|---|
| Structured signal→strategy→outcome | Yes | No (free-form text) | No | No |
| Multi-agent cross-vendor sharing | Yes | No | No | No |
| Human-in-the-loop promotion | Yes | No | Limited | No |
| MCP native (zero adapter code) | Yes | No | No | No |
| Confidence decay + auto-archival | Yes | No | No | No |
| Zero external dependencies (no Docker/Postgres/Neo4j) | Yes | No | No | No |
Contributing
See CONTRIBUTING.md.
License
MIT — see LICENSE.
Python SDK coming soon.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.