API Documentation

Free REST API for querying 5,900+ AI agent capabilities across 7 platforms. No authentication required. JSON responses. CORS enabled.

Base URL
https://agentsift.ai
Auth
None required
Rate Limit
100 req/min
Format
JSON

Quick Start

Search for file system capabilities on MCP:

curl "https://agentsift.ai/api/capabilities?platform=mcp&category=file-system&limit=5"

Semantic search — describe what you need in plain English:

curl "https://agentsift.ai/api/search/semantic?q=send+email+safely&limit=5"

Compare capabilities side by side:

curl "https://agentsift.ai/api/compare?ids=mcp-filesystem-read-file,openclaw-read"

SDKs

Official clients for Python and TypeScript. Zero runtime dependencies. Free forever.

🐍

Python SDK

stdlib only
pip install agentsift
from agentsift import AgentSift

client = AgentSift()

# List capabilities
caps = client.capabilities.list(platform="mcp", limit=5)
for cap in caps:
    print(cap.name, cap.risk_level)

# Semantic search
results = client.search.semantic("send email safely")

# Compare
comp = client.compare.compare(
    "mcp-filesystem-read-file",
    "openclaw-read"
)
print(comp.verdict)
GitHub → sdk/python
📦

TypeScript SDK

native fetch
npm install agentsift
import { AgentSift } from 'agentsift';

const client = new AgentSift();

// List capabilities
const caps = await client.capabilities.list({
  platform: 'mcp', limit: 5,
});
caps.forEach(cap => console.log(cap.name, cap.riskLevel));

// Semantic search
const results = await client.search.semantic('send email safely');

// Compare
const comp = await client.compare(
  'mcp-filesystem-read-file',
  'openclaw-read',
);
console.log(comp.verdict);
GitHub → sdk/typescript
Requirements: Python ≥ 3.8 (no deps) · Node ≥ 18 (native fetch) · Works in Deno + modern browsers

Endpoints

Response Format

All endpoints return JSON. Capability objects include:

{
  "id": "mcp-filesystem-read-file",
  "name": "Read File",
  "platform": "mcp",
  "category": "file-system",
  "description": "Read contents of a file from the local filesystem",
  "riskLevel": "caution",
  "trustScore": 0.72,
  "permissions": ["filesystem.read"],
  "tags": ["file", "read", "filesystem", "io"],
  "source": "https://github.com/modelcontextprotocol/servers"
}

Use Cases

Agent Builder Tools

Let users discover and compose capabilities for custom agents

Security Audits

Assess permission surfaces and risk levels before deploying agent stacks

Platform Migration

Find equivalent capabilities when switching between agent frameworks

Capability Monitoring

Track new additions via RSS feeds and the changelog API

Error Codes

StatusMeaningCommon Cause
200OKRequest succeeded
400Bad RequestMissing required params, invalid filter value, or malformed query
404Not FoundCapability ID doesn't exist in the registry
422UnprocessableCompare endpoint: need 2-3 valid IDs, got <2 or >3
429Rate LimitedOver 100 requests/min — back off and retry with exponential delay
500Server ErrorBug on our end — report to @OnlyAMicrowave
// Error response format
{
  "error": "Invalid platform filter",
  "status": 400
}

Rate Limiting

All endpoints share a pool of 100 requests per minute per IP. Exceeding this returns 429 with a Retry-After header (seconds).

Tips for high-volume usage

  • Use limit=200 to reduce total requests when paginating
  • Cache responses — data refreshes nightly at 4 AM ET, not in real-time
  • Use the RSS feed instead of polling for new capabilities
  • Need higher limits? Reach out via Twitter DM

Built by @OnlyAMicrowave · Updated nightly at 4 AM ET