Empower AI Agents with real-world services.

Search, connect, and invoke data, SaaS, and expert capabilities through one unified service layer with 1,092 capabilities available through API, SDK, or MCP.

Explore APIs Start Invoking

APIs & Services · workflow

01Identify the missing capability

02Connect it to the workflow

03Make it operational

Service connections: 100+
Callable actions: 500+
Unified interface & billing: 1

Choose capabilities first, then invocation methods.

Start with real SandBase capabilities, switch between MCP, CLI, API, SDKs, and Workflow Skills, and immediately view real requests and structured results.

6 capability groups · 1,092+ callable capabilities

Claude · ChatGPT · Codex

REQUEST · MCP

web.search · MCP

{
  "mcpServers": { "sandbase": { "command": "npx", "args": ["-y", "@sandbase/mcp"], "env": { "SANDBASE_API_KEY": "••••••••" } } }
}

Tool: web.search

Credentials do not enter prompt or model context.

RESPONSE · JSON · 200 OK · 684ms

{
  "success": true,
  "capability": "web.search",
  "provider": "Exa · Tavily · Firecrawl",
  "results": 5,
  "trace_id": "run_8f21c4",
  "data": [
    {
      "title": "Agent infrastructure",
      "score": 0.97
    },
    {
      "title": "Production AI systems",
      "score": 0.94
    }
  ]
}

Connect SandBase to each AI Agent.

Bring the same set of SandBase capabilities into any AI Agent, MCP client, product, or backend.

MCP · Connect AI tools

Let Claude, ChatGPT, Codex, and any MCP client discover and invoke SandBase capabilities.

npx -y @sandbase/mcp connect

SDK · Configure capabilities

Embed SandBase calls into your Agent runtime using Node.js, Python, Go, or Rust.

await sandbase.run("web.search", input)

API · Integrate a product backend

Call one unified endpoint from any HTTP application and receive structured results plus an execution trace.

POST /v1/capabilities/run

Connections have boundaries, actions are recorded.

Each user has an independent connection. Permission scope, request parameters, duration, results, and costs are included in the same execution trace.

✓ Credentials do not enter prompt or model context
✓ OAuth token auto-refresh and expiry check
✓ Limit permission scope by service and action

HubSpot · search_contacts

12 records · 684ms · $0.02

Notion · create_pages

12 pages · 1.2s · $0.04

Slack · send_message

1 message · 421ms · $0.01

Total · 2.3 seconds

Explore APIs Read documentation

From selecting a service to performing the first real operation.

Choose the exact action, connect the user account with a limited permission scope, and invoke it through the interface that fits your product or Agent.

01
Choose service and action
Start from a task and find the capability to search, read, write, or notify.
Search · Select action
02
Connect account and limit permissions
Use managed authorization to choose the actions each user is allowed to perform.
Connect · Per-user scope
03
Choose API, SDK, or MCP access
Call it directly in your product or configure it as a tool for your Agent.
API · SDK · MCP

Create API key Build an Agent

Turn a useful capability into repeatable work.

Connect it to the tools your team uses now, or compose it into an Agent with a clear job.

Use in AI Tools Build an Agent