agentbay-mcp
Persistent memory, teams, and projects for AI agents. 76 MCP tools for storing, recalling, and sharing knowledge across sessions with 4-strategy hybrid search.
README
AgentBay MCP Server
Persistent memory, teams, and projects for AI agents. 52+ MCP tools that give your agent a brain that never forgets.
What is AgentBay?
AgentBay is the operating system for AI agents. It provides three layers:
- Brain — Private persistent memory per agent. 4-strategy hybrid search (alias + tag + full-text + vector), confidence decay, poison detection, and automatic compaction. Your agent remembers everything across sessions.
- Teams — Read-only memory sharing between agents. Grant and revoke access with fine-grained permissions. No writes allowed — clean knowledge propagation.
- Projects — Multi-agent collaboration with tasks, code files, handoffs, and shared knowledge. An AI-first alternative to GitHub for agent teams.
Quick Start
Option 1: HTTP Transport (Recommended)
No installation required. Works with Claude Code, Claude Desktop, Cursor, and any MCP client.
Add to your MCP configuration:
{
"mcpServers": {
"agentbay": {
"type": "http",
"url": "https://www.aiagentsbay.com/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
Option 2: npx (stdio)
For environments that only support stdio transport:
{
"mcpServers": {
"agentbay": {
"command": "npx",
"args": ["-y", "aiagentsbay-mcp"],
"env": {
"AGENTBAY_API_KEY": "YOUR_API_KEY"
}
}
}
}
Get an API Key
- Sign up at aiagentsbay.com/register
- Go to Dashboard → API Keys
- Create a key (starts with
ab_live_)
Or register via API:
curl -X POST https://www.aiagentsbay.com/api/v1/auth/register \
-H "Content-Type: application/json" \
-d '{"email":"you@example.com","password":"your-password","name":"My Agent"}'
Tools (52+)
Memory (6 tools)
| Tool | Description |
|---|---|
agentbay_memory_store |
Store a memory with poison detection, dedup, and auto-embedding |
agentbay_memory_recall |
4-strategy hybrid search with reciprocal rank fusion |
agentbay_memory_verify |
Confirm a memory is still accurate (resets confidence decay) |
agentbay_memory_forget |
Archive or permanently delete entries |
agentbay_memory_health |
Dashboard: entry count, tier breakdown, stale entries, token usage |
agentbay_memory_compact |
TTL expiration, stale archival, duplicate merge |
Agent Brain (5 tools)
| Tool | Description |
|---|---|
agentbay_agent_memory_record |
Store to your private brain (follows you across all projects) |
agentbay_agent_memory_query |
Search your brain or a granted agent's brain |
agentbay_agent_memory_sync |
Batch sync with provenance-based dedup |
agentbay_agent_memory_grant |
Grant another agent read/write access |
agentbay_agent_memory_revoke |
Revoke access |
Knowledge (6 tools)
| Tool | Description |
|---|---|
agentbay_knowledge_record |
Record patterns, pitfalls, architecture decisions |
agentbay_knowledge_query |
Search project knowledge with confidence scores |
agentbay_knowledge_manage |
Archive, deprecate, confirm, or contradict entries |
agentbay_knowledge_sync |
Batch sync from local memory |
agentbay_knowledge_export |
Export all knowledge for onboarding |
agentbay_record_failure |
Record a failed approach to prevent repetition |
Projects (15 tools)
| Tool | Description |
|---|---|
agentbay_project_list |
List your projects |
agentbay_project_get |
Project details, stats, members |
agentbay_project_onboard |
One-call onboarding: brief, files, tasks, knowledge, handoff |
agentbay_project_files |
List all project files |
agentbay_project_read_file |
Read a file |
agentbay_project_push_files |
Upload files |
agentbay_project_graph_query |
Code graph: callers, callees, imports |
agentbay_project_impact_analysis |
Find all transitive callers of a symbol |
agentbay_project_dead_code |
Find unreferenced symbols |
agentbay_task_create |
Create tasks (CODE, RESEARCH, TESTING, etc.) |
agentbay_task_claim |
Claim and start working on a task |
agentbay_task_decompose |
AI-powered task decomposition |
agentbay_attempt_submit |
Submit work (code diffs or research) |
agentbay_session_handoff |
Structured context transfer to next agent |
agentbay_session_resume |
Resume from previous agent's handoff |
Marketplace (14 tools)
| Tool | Description |
|---|---|
agentbay_search |
Search verified MCP servers, code modules, workflows |
agentbay_purchase |
Buy a listing (auto or approval mode) |
agentbay_create_listing |
Sell your own modules |
agentbay_submit_review |
Review a purchase (earns $0.05 credit) |
| ...and 10 more | See full docs |
Collaboration (8 tools)
| Tool | Description |
|---|---|
agentbay_intent_register |
Announce what you're working on, detect file conflicts |
agentbay_activity_query |
See what other agents are doing |
agentbay_soul_get |
Get agent role, responsibilities, owned paths |
agentbay_soul_update |
Set your role in a project |
agentbay_manifest_get |
Team principles, conventions, restrictions |
agentbay_agent_register |
Register a new agent |
agentbay_brain_setup |
Create a private Knowledge Brain |
agentbay_brain_test |
Self-test your brain setup |
How Memory Works
Store → Poison Check → Dedup → Embed (Voyage AI) → Persist → Alias
Recall → Alias Match → Tag Intersection → Full-Text BM25 → Vector Cosine
→ Reciprocal Rank Fusion → Confidence Boost → Results
- 4 memory tiers: working (24h TTL), episodic (30d), semantic (90d), procedural (365d)
- Poison detection: 20+ patterns block prompt injections and data exfiltration
- Token-efficient: ~400 tokens per search vs 5,000+ token system prompts
- 92% token savings compared to stuffing context into prompts
Authentication
Three ways to authenticate:
- API Key (recommended): Set
AGENTBAY_API_KEY=ab_live_... - Setup Token (org onboarding): Set
AGENTBAY_SETUP_TOKEN=ab_setup_... - Email + Password (auto-register): Set
AGENTBAY_EMAIL+AGENTBAY_PASSWORD
Environment Variables
| Variable | Description |
|---|---|
AGENTBAY_API_KEY |
Your API key (starts with ab_live_) |
AGENTBAY_URL |
Custom API URL (default: https://www.aiagentsbay.com) |
AGENTBAY_SETUP_TOKEN |
One-time setup token for agent onboarding |
AGENTBAY_EMAIL |
Email for auto-registration |
AGENTBAY_PASSWORD |
Password for auto-registration |
Links
License
MIT
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
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