LangChain tools for
632 live web-data APIs

Install and set your key

Install the package, then put your ReefAPI key in the environment. You can also pass it to the constructor, ReefAPIToolkit(api_key="..."). The key is held as a SecretStr and left out of repr(), logs and serialized tool state. The two discovery tools work without a key.

pip install -U langchain-reefapi

export REEFAPI_API_KEY="your-key"     # REEFAPI_KEY works too

No key yet? Create a free account for 1,000 credits, no card.

Three tools and a toolkit

ReefAPIToolkit returns all three with one shared configuration: ReefAPIToolkit().get_tools(). The tools are built on the reefapi Python SDK, so retries for retryable errors and the client-side rate limit (30 requests per 3 seconds per key) come with them. Every tool has a sync and an async path.

An agent that picks the engine itself

Any chat model with tool calling works; this one uses Anthropic through langchain[anthropic].

pip install -U langchain-reefapi "langchain[anthropic]"

from langchain.agents import create_agent
from langchain_reefapi import ReefAPIToolkit

agent = create_agent(
    model="claude-sonnet-4-6",          # any chat model with tool calling
    tools=ReefAPIToolkit().get_tools(),
    system_prompt="Use the ReefAPI tools for live data from websites. "
                  "Find the engine first, read its schema, then call it.",
)

result = agent.invoke(
    {"messages": [{"role": "user", "content": "What are the 3 cheapest iPhone 13 Pro listings on eBay right now?"}]}
)
print(result["messages"][-1].content)

A typical run: the model calls reefapi_search_engines (for example with “ebay listings”), then reefapi_get_schema("ebay", "search"), then reefapi_call with the parameters it read.

The tools one by one

Steps 1 and 2 need no key. The same parameters are on each engine's docs page, for example /docs/ebay.

import json
from langchain_reefapi import ReefAPICall, ReefAPIGetSchema, ReefAPISearchEngines

# 1. which engine? (no key needed)
found = json.loads(ReefAPISearchEngines().invoke({"query": "sold prices on ebay", "max_results": 3}))
print([e["engine"] for e in found["engines"]])          # ['ebay', 'cardladder', 'ebay-sold']

# 2. which parameters? (no key needed)
schema = json.loads(ReefAPIGetSchema().invoke({"engine": "ebay", "action": "search"}))
print(schema["required_params"], schema["example_params"])   # ['query'] {'query': 'iphone 13 pro'}

# 3. call it (uses your key)
data = json.loads(ReefAPICall().invoke(
    {"engine": "ebay", "action": "search", "params": {"query": "iphone 13 pro", "max_results": 5}}
))
for item in data["results"]:
    print(item["title"], item["price"]["value"], item["price"].get("currency"))

Errors go back to the agent; failures cost nothing

Every error becomes a ToolException whose text carries the error code (INVALID_PARAM, UNKNOWN_API, RATE_LIMITED, QUOTA_EXCEEDED, TARGET_BLOCKED, UPSTREAM_TIMEOUT, ...), whether a retry can help, the HTTP status and the request id:

ReefAPI error INVALID_PARAM: sort must be one of ... retryable=false. http_status=400. request_id=...
Check the parameter names and allowed values with reefapi_get_schema. Failed calls are not charged.

Set handle_tool_error=False to get the exception instead; it has .code, .retryable and .request_id. Plans and credit prices: pricing.

Constructor arguments

All three tools and the toolkit accept:

Not using LangChain?