AgentDB

Real-time curated knowledge API for AI agents.

AgentDB is a knowledge base updated Mon/Wed/Fri with summaries from 14 sources spanning markets, technology, science, philosophy, and current events. Connect it to your AI agent so it always has fresh context — without you having to manage scraping, summarisation, or storage.

---

Quick start

1. Get an API key

curl -s -X POST https://agentdb-production-9ba0.up.railway.app/v1/auth/register \
  -H "Content-Type: application/json" \
  -d '{"email": "you@example.com", "name": "Your Agent"}'

Response:

{
  "api_key": "adb_xxxxxxxxxxxxxxxxxxxx",
  "tier": "trial",
  "trial_expires_at": "2026-04-21T07:00:00",
  "message": "Welcome to AgentDB. Your 3-day trial has started."
}

Store your key — it's shown once.

2. Fetch the latest knowledge

curl https://agentdb-production-9ba0.up.railway.app/v1/knowledge/latest \
  -H "X-API-Key: adb_xxxxxxxxxxxxxxxxxxxx"

---

MCP server (recommended for Claude)

The AgentDB MCP server exposes two tools directly inside Claude Code or Claude Desktop: get_latest_knowledge and search_knowledge.

Install with uv (recommended)

uv handles Python version management automatically — no need to install Python 3.10+ manually.

curl -LsSf https://astral.sh/uv/install.sh | sh

Claude Code — global config

Add to ~/.claude.json (works in every Claude Code session):

{
  "mcpServers": {
    "agentdb": {
      "type": "stdio",
      "command": "/path/to/uv",
      "args": ["run", "/path/to/agentdb/mcp/server.py"],
      "env": {
        "AGENTDB_API_KEY": "adb_xxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Find your uv path with which uv. Find the server path with realpath mcp/server.py from this repo.

Claude Code — project config

Add a .mcp.json to your project root:

{
  "mcpServers": {
    "agentdb": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "/absolute/path/to/agentdb/mcp/server.py"],
      "env": {
        "AGENTDB_API_KEY": "adb_xxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "agentdb": {
      "command": "uv",
      "args": ["run", "/absolute/path/to/agentdb/mcp/server.py"],
      "env": {
        "AGENTDB_API_KEY": "adb_xxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Run the MCP server manually (test it)

AGENTDB_API_KEY=adb_xxxxxxxxxxxxxxxxxxxx uv run mcp/server.py

Available MCP tools

Tool	Tier	Description
get_latest_knowledge	Trial + Pro	Fetch the N most recent items, optionally filtered by tags or content type
search_knowledge	Pro	Semantic vector search — find items most relevant to a natural language query

---

REST API reference

Base URL: https://agentdb-production-9ba0.up.railway.app

All endpoints (except /health, /v1/auth/register, and /v1/knowledge/sources) require:

X-API-Key: adb_xxxxxxxxxxxxxxxxxxxx

Knowledge

Method	Path	Tier	Description
GET	/v1/knowledge/latest	Trial + Pro	Latest items; supports limit, page, tags, content_type
GET	/v1/knowledge/search?q=...	Pro	Semantic search
GET	/v1/knowledge/{id}	Trial + Pro	Single item by ID
GET	/v1/knowledge/sources	Public	List of all scraped sources

Query parameters — /latest

Param	Type	Default	Description
limit	int	20	Items per page (1–100)
page	int	1	Page number
tags	string	—	Comma-separated tag filter, e.g. ai,markets
content_type	string	—	article, video, research, or data

Example response item

{
  "id": "3e7c224c-...",
  "title": "Quantum Jamming and the Search for Principles Deeper Than Quantum Mechanics",
  "content_type": "article",
  "summary": "Researchers are exploring whether cryptographic protocols...",
  "body": {
    "category": "science_research",
    "key_points": ["...", "..."],
    "source_name": "Quanta Magazine"
  },
  "tags": ["quantum-mechanics", "cryptography", "causality"],
  "confidence": 0.92,
  "relevance_score": 0.92,
  "published_at": "2026-04-18T11:40:55+00:00"
}

Auth

Method	Path	Description
POST	/v1/auth/register	Register and get a trial API key
GET	/v1/auth/me	Inspect your key (tier, usage, expiry)

---

Sources

AgentDB ingests from 14 sources, updated Mon/Wed/Fri at 07:00 UTC.

YouTube channels (video summaries)

Source	Category
Bloomberg	market_news
CNBC	market_news
Reuters	market_news
Prof G Markets	market_news
Rebel Capitalist	market_news_alternative
Lex Fridman	technology_ai
Y Combinator	startups_technology
Closer To Truth	philosophy_science
Bernardo Kastrup	philosophy_science

Blogs / RSS (article summaries)

Source	Category
Hacker News	technology_startups
Ars Technica	technology_science
Quanta Magazine	science_research
Marginal Revolution	economics_policy
Zero Hedge	market_news_alternative

Full machine-readable list: GET /v1/knowledge/sources

---

Pricing

Tier	Price	Rate limit	Features
Trial	Free	100 req/day	Latest items, 3 days
Pro	$20/month	1,000 req/day	Latest + semantic search
Fleet	$99/month	10,000 req/day	Everything, bulk access

Upgrade via /v1/payments/checkout (Stripe or crypto).

---

Self-hosting

git clone https://github.com/AgentBC9000/agentdb
cd agentdb
cp .env.example .env   # fill in ANTHROPIC_API_KEY, DATABASE_URL, ADMIN_SECRET
docker-compose up

The scraper runs independently — trigger it manually or set a cron:

cd scraper
ANTHROPIC_API_KEY=... AGENTDB_API_URL=... ADMIN_SECRET=... python run.py