Real-time data + action layer

The real-time data and action layer for AI agents.

One endpoint where agents search, verify, and act on real businesses — live availability, prices, and facts. Ranked, structured, and fresher every time an agent calls.

View docs

Restaurants · Salons · Fitness · Hotels · Home services

loop.search() · liveself-correcting

// agent → loop · one tool call
search(
  "quiet outdoor table for 4, vegan options",
  "Kreuzberg, Berlin"
)

// 2 results · ranked · machine-readable
[
  {
    "name":         "Café Botanika",
    "availability": "likely_free_tonight",
    "confidence":   0.92,
    "vegan":        true,
    "price_band":   "€€",
    "observed":     "2h ago"  fresh
  },
  {
    "name":         "Hartmann's Garden",
    "availability": "call_to_confirm",
    "confidence":   0.74,
    "observed":     "5h ago"
  }
]

§ 01 How it works

Four tools. One endpoint.

Natural language in. Ranked, structured answers out. Built for tool calls, not human browsing.

search(query, location, filters)

Find, ranked

Natural-language query in — ranked, machine-readable results out, scored for fit.

→ returns ranked[] with confidence

get_details(result_id)

Full record

The complete structured record for a chosen result — hours, menu, price, contact.

→ returns normalized schema

verify(result_id, claim)

Freshness check

Confirm a single fact before acting — the latest observation and exactly when it was seen.

→ returns value · observed_at

report(result_token, outcome)

Close the loop

Tell Loop what happened. Agents that report outcomes earn fresher data and higher limits.

→ returns updated limits

§ 02 The loop

Every agent call makes the data better.

Demand, selection, and outcome signals flow back from normal usage. The data layer corrects itself — no scraping cron, no stale cache.

More agents → better data → more agents

1Call

An agent asks in natural language — find, check, decide.

search() · verify()

2Answer

Ranked, structured results — each stamped with when it was seen.

confidence · observed

3Act

The agent books, reserves, or buys in the real world.

get_details()

4Signal

Outcome flows back. The data underneath corrects itself.

report()

The loopEvery call sharpens the dataMore agents → better data → more agents

§ 03 Why Loop

Built for agents, not maps.

Map APIs confirm a place exists. Agents need to know what’s true right now — in a shape they can act on.

Dimension

Built for humans & maps

Loop · built for agents

Freshnessis it true now?

Humans & mapsStatic records. You learn it’s wrong after the call fails.

LoopEvery field stamped observed 2h ago. Verify before you act.

Structurecan an agent use it?

Humans & mapsHTML and prose meant for eyes. Burn tokens parsing it.

LoopRanked JSON with confidence scores. Drop straight into a tool call.

Coverageper integration

Humans & mapsOne source per vertical. Wire up and maintain hundreds.

LoopOne normalized schema across every vertical. One integration.

§ 04 Integration

Works where your agent lives.

Add Loop as an MCP server in one config line, or call the REST endpoint directly. Same data, same freshness, either way.

Loop speaks the protocols your stack already uses — drop it into any agent framework as a tool.

MCPClaude & any MCP-compatible client

REST OpenAI tools, function calling

SDK LangChain-style agent frameworks

Named as contexts, not partnerships.

~/.claude/mcp.json

{
  "mcpServers": {
    "loop": {
      "command": "npx",
      "args": ["-y", "@loop/mcp"],
      "env": { "LOOP_API_KEY": "sk_live_…" }
    }
  }
}

curl · api.stayinloop.dev

curl https://api.stayinloop.dev/v1/search \
  -H "Authorization: Bearer $LOOP_API_KEY" \
  -G \
  --data-urlencode 'q=quiet vegan table for 4 tonight' \
  --data-urlencode 'location=Kreuzberg, Berlin'

# → 200 OK · ranked results, each with observed_at

§ 05 Pricing

Start free. Scale when your agents do.

Usage-based, no seats. Reporting outcomes earns fresher data and higher limits on every tier.

Free

1,000queries / mo

Wire Loop into your agent and ship a prototype.

All four tools
MCP server + REST
Community support

Pro

Usage-basedplaceholder

Scales with calls. For agents in production.

Higher rate limits
Priority freshness windows
Report → reward limits
Email support

Scale

Talk to uscustom

High-volume agents and dedicated verticals.

Custom limits & SLA
Private endpoints
Vertical onboarding