langchain-dexpaprika

DexPaprika tools for LangChain agents. No API key and no signup to start: the DexPaprika API free tier is keyless, so your agent gets DEX market data across every supported blockchain (36 chains, 33M+ tokens, 36M+ pools) as soon as you install the package.

We built these tools for LLM consumption: descriptions tell the model exactly which parameters exist and where to get their values, error messages quote the API's own allowed-value lists so the model can self-correct, and outputs are compact JSON with multi-kilobyte token descriptions trimmed.

Installation

pip install -U langchain-dexpaprika

No credentials to configure. This package calls the keyless free tier and sends no API key, so there is no environment variable to set. The free tier is keyless, with data delayed up to 15 seconds; see https://docs.dexpaprika.com/knowledge-base/rate-limits for the current rates and quotas, and https://dexpaprika.com/api/pricing for the paid tiers.

Quickstart

from langchain_dexpaprika import DexPaprikaSearch

search = DexPaprikaSearch()
print(search.invoke({"query": "WETH"}))

This returns compact JSON with matching tokens (contract address, chain, USD price, liquidity), pools, and DEXes.

Tools

Use the toolkit in an agent

DexPaprikaToolkit bundles all five tools over one shared HTTP client. The example below drives them with an Anthropic model, so install the provider and set its key first (swap in any chat model you prefer):

pip install -U "langchain[anthropic]"
export ANTHROPIC_API_KEY=...

from langchain.agents import create_agent
from langchain_dexpaprika import DexPaprikaToolkit

toolkit = DexPaprikaToolkit()
agent = create_agent("claude-sonnet-4-5", toolkit.get_tools())
result = agent.invoke(
    {"messages": [("user", "Find the most liquid WETH pool on ethereum")]}
)

The tool descriptions chain naturally: an agent starts with dexpaprika_search to resolve a ticker into a contract address and network id, then feeds those into the other tools.

Individual tools

Every tool works standalone, sync and async:

from langchain_dexpaprika import DexPaprikaPoolOHLCV

ohlcv = DexPaprikaPoolOHLCV()
candles = ohlcv.invoke(
    {
        "network": "ethereum",
        "pool_address": "0x88e6a0c2ddd26feeb64f039a2c41296fcb3f5640",
        "start": "-24h",  # the last 24 hours, which works without a key
        "interval": "1h",
        "limit": 24,
    }
)

start and end take an offset back from now (-24h, -7d, -90m) as well as YYYY-MM-DD, RFC3339 and Unix seconds. How far back you can go and how fine the candles can be depends on your plan: without a key, the last 24 hours at 1h and longer; a free key opens 7 days at 10m and longer; Dev 30 days at every interval; Pro unlimited. A request outside your plan fails with the API's message, which names the plan that allows it. Full table: OHLCV limits by plan.

Using an API key (optional)

The toolkit works without a key. No signup, no card.

A free key raises the monthly credit allowance and the per-minute request limit, and opens 7 days of OHLCV history. Current figures are on the rate limits page.

from langchain_dexpaprika import DexPaprikaToolkit
from langchain_dexpaprika._client import DexPaprikaAPIWrapper

toolkit = DexPaprikaToolkit(api_wrapper=DexPaprikaAPIWrapper(api_key="api_your_key"))

# Or leave it out and set DEXPAPRIKA_API_KEY in the environment
toolkit = DexPaprikaToolkit()

An explicit api_key beats the environment variable, and no key at all keeps the previous keyless behaviour unchanged.

The key is excluded from repr() and model_dump(), because these wrappers end up inside agent traces and serialized chains, and a key in a trace is a credential in somebody's logs.

There is no Bearer prefix. The key is sent as the entire Authorization value, which is what the API expects; ApiKey or Token in front of it returns 401.

Pro customers also set base_url to https://api-pro.dexpaprika.com. The host never changes on its own, because a free key sent to the Pro host returns 403.

Error handling

We surface API errors as messages the agent can act on:

• 400 responses quote the API's message verbatim, including the exact list of allowed values for the offending parameter.
• 404 responses arrive with an empty body, so we synthesize a message that names the missing identifier and points at the tool that lists valid values.
• 429 responses are retried automatically, honoring the Retry-After header, before we tell the agent to slow down.
• 410 responses (removed endpoints) surface the replacement endpoint the API reports, with a hint to upgrade the package.

Development

uv venv
uv pip install -e '.[test]'
make lint                # ruff + mypy
make test                # unit tests, sockets blocked
make integration_tests   # live API, keyless, run serially

Links

• DexPaprika API documentation
• DexPaprika agent integration guide
• API reference

License

MIT