langchain-stackresolve

LangChain tools for StackResolve: web intelligence for AI agents.

Pick software for a task, compare vendors, check whether a product is agent-ready, and pull structured company facts. One tool call instead of a search-and-scrape loop.

pip install langchain-stackresolve

Quickstart

Every tool works without a key, subject to an anonymous rate limit. A free key from stackresolve.dev/developers raises it.

import os
from langchain_stackresolve import StackResolveToolkit
from langchain.agents import create_agent

os.environ["STACKRESOLVE_API_KEY"] = "ar_..."  # optional, raises the rate limit

agent = create_agent(
    model="claude-sonnet-5",
    tools=StackResolveToolkit().get_tools(),
)

result = agent.invoke({
    "messages": [{
        "role": "user",
        "content": "I need to scrape javascript-heavy sites. What should I use, "
                   "and what does it cost?",
    }]
})
print(result["messages"][-1].content)

The agent calls stackresolve_find_tools to get scored candidates, then stackresolve_get_pricing on the winner. Two calls, structured answers, no scraping.

Single tools

Import only what you need:

from langchain_stackresolve import StackResolveAudit, StackResolveFindTools

audit = StackResolveAudit()
print(audit.invoke({"domain": "stripe.com"}))

find = StackResolveFindTools()
print(find.invoke({"task": "send transactional email from a Node service"}))

Tools

All eight work without a key, subject to an anonymous rate limit. A free key raises the limit and is required for account endpoints (monitors, usage, discovery runs), which this package does not expose.

Configuration

The toolkit and every tool accept api_key and base_url, and otherwise read STACKRESOLVE_API_KEY and STACKRESOLVE_BASE_URL from the environment.

toolkit = StackResolveToolkit(api_key="ar_...")

Pass an existing SDK client to share connection state:

from stackresolve import StackResolve

client = StackResolve(api_key="ar_...")
tools = StackResolveToolkit(client=client).get_tools()

Errors

Tools return a readable message rather than raising, so a failed call does not end the agent's turn. A missing key, an exhausted allowance, and a rate limit each come back as text the model can act on.

Output size

Tool output is capped at 6,000 characters and marked when truncated, so a large registry or research payload cannot flood the context window. Change it with langchain_stackresolve.tools.MAX_CHARS.

Also available

• Hosted MCP server (no install): https://mcp.stackresolve.dev/mcp
• TypeScript SDK and CLI: npm install stackresolve
• Python SDK on its own: pip install stackresolve
• REST API: https://api.stackresolve.dev, OpenAPI at /openapi.json

License

MIT