Langchain KeeperHub

langchain-keeperhub connects LangChain/LangGraph agents to KeeperHub so agents can transfer tokens, call contracts, conditionally execute, resolve ENS, and track txs via LangChain tools—using only a KeeperHub API key, with no private keys in app code (signing via KeeperHub / Turnkey TEE). Includes safety options, optional execution history, direct client usage, and MCP workflows. Open-source Python SDK on PyPI for agentic Web3 without key management in the agent stack.

I used Python, httpx for async HTTP to KeeperHub’s API, and pydantic + langchain-core so everything fits normal LangChain / LangGraph apps. The toolkit is the agent-facing layer; KeeperHubClient is the same API in a direct, script-friendly form (async async with style) so I’m not forcing everyone to run an agent.

ENS I kept on public JSON-RPC with simple eth_call patterns so I could skip web3.py. For the sync tool surface vs async client, I added a small async bridge (asyncio.run, or a thread when a loop is already running in notebooks or web apps).

For KeeperHub MCP, I load their hosted server when workflows are enabled, rename colliding tools (e.g. with a workflow_ prefix) so names stay clear, and strip noisy JSON Schema metadata so model providers like Gemini bind tools cleanly.

The ideas I cared about most: no write auto-retries (avoid double sends), guardrails before the network, and one codebase that works for “agent with tools” and “small Python automation with the client.”