
hunterkiller
io.github.Irishman-Management-LLC/hunterkiller
Liquidation walls with snapshot age and evidence labels; keyless public MCP discovery and Free data.
Documentation
Hunter Killer agent examples
curl --fail-with-body --max-time 20 https://hunterkiller.io/api/v1/levels/BTCUSDT
# Or, with only Python's standard library:
python3 examples/one_call.py BTCUSDT
The public keyless door is live as of release 148 on 2026-11-07. The first call needs no key for the six Free coins, 2 hours behind. Wrapper publication is a separate step; no external paying use is claimed. A failed call remains an error.
Read liquidation walls, snapshot age, evidence labels and a public record. The strategy examples use paper orders. Live execution is disabled. Descriptive measurements, not trading advice.
Free data is 2 hours behind on six coins. Inspect tier, delay_sec,
as_of_iso, age_sec, nearest, walls, truth and record. Keep these
fields with any result you show. An empty ladder is not a trading instruction.
Calls can return 401, 403, 402, 429 or 503; errors must reach the caller.
Internal verification must use python3 examples/one_call.py --probe so probes
are not confused with external adoption.
Integrations
packages/langchain:hunterkiller-langchain, Python transport and LangChain factory.packages/crewai:hunterkiller-crewai, CrewAI factory using the same transport.packages/composio:hunterkiller-composio, experimental standalone session tool.packages/npm:hunterkiller-agent-tools, Node client, LangChain factory and function definitions.schemas/openai-function.json: strict Responses API function schema. Dispatch to the read only HTTP client.schemas/composio-openapi.json: public REST definition for custom bindings, not a hosted catalog listing.space: manual Gradio ladder demo for Hugging Face Spaces.examples/sdk.pyandexamples/sdk.mjs: existing published SDK Desk Read examples.examples/freqtrade: the Freqtrade paper strategy, native historical public wall proxy table and API key signup link.examples/hummingbot: the Hummingbot paper grid and its labeled historical quote model table.strategies: shared point in time inputs, sealed observations, full results, costs and losses.
The wrappers call the public endpoint directly because the versions of the
existing SDKs currently on PyPI and npm do not expose levels. They do not
compute walls or grades. Credentials, if supplied by the application, never
appear in the model's argument schema. No payment is automatic.
MCP
https://hunterkiller.io/mcp is the public door. Discovery, list_symbols
and hk_levels need no key for the six Free coins, 2 hours behind. REST and
these two tools share 60 requests per minute and 500 per UTC day per IP.
Response headers state the actual limit and quota mode. Directory health
checks that do not authenticate use this public apex URL.
https://agents.hunterkiller.io/mcp is the Managed OAuth transport for live
tools. Complete its normal sign in flow in your MCP client. OAuth transport
access and paid Hunter Killer data entitlement are separate. The protected
apex twin /mcp/protected retains the same gate. registry/server.json
version 1.0.2 lists only the public apex so unauthenticated directory checks
select the public door. The OAuth transport remains available separately.
The registry record is hosted metadata, not an installable open source server
implementation.
Public source and risk
Agent door, proof, privacy, risk. Never infer a future return or trade recommendation from these measurements. The paper examples retain their historical proxy inputs, costs and losing periods. Those results describe a simulation, not live performance.
No installation packages available.