Next free-source huntWe check for new free data every weekday.

Integrations · Python

Signal Lab in LangChain and LangGraph

Load the Signal Lab tools into a LangChain agent (built on LangGraph) with the MCP adapter, then call them like any tool.
https://signal-lab.qdt.ai/api/mcp

Set it up

  1. The MCP adapter connects over Streamable HTTP and turns each Signal Lab tool into a LangChain tool.
  2. create_agent builds a LangGraph agent, so the same tools work in a custom LangGraph graph.
  3. On an older stack, langchain-mcp-adapters (MultiServerMCPClient) loads the same tools; see the second snippet.
bash
pip install "langchain[mcp]"
agent.py
import asyncio
from langchain.agents import create_agent
from langchain.mcp import MCPAdapter

MODEL = "provider:model-name"  # any chat model LangChain supports

async def main():
    async with MCPAdapter("https://signal-lab.qdt.ai/api/mcp") as adapter:
        tools = await adapter.list_tools()
        agent = create_agent(MODEL, tools)
        result = await agent.ainvoke(
            {"messages": [{"role": "user", "content": "Find free daily FX rate sources with an API and no key, then show me the ingestion skill for the closest match."}]}
        )
        print(result["messages"][-1].content)

asyncio.run(main())
agent_adapters.py (langchain-mcp-adapters)
from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient(
    {"signal_lab": {"transport": "http", "url": "https://signal-lab.qdt.ai/api/mcp"}}
)
tools = await client.get_tools()  # in an async function; pass to create_agent or a LangGraph ToolNode

What your agent gets

  • check_stack: Review an MCP client config (Claude Desktop, Cursor or VS Code JSON): deprecated or stale servers with replacements, estimated tool count and token cost, Cursor's 40-tool cap, and risky tool mixes. Parsed only; never run or stored. Agent Stack Doctor
  • get_trust_score: The Trust Consensus score (0-100, A-E) for one audited MCP server: our audit and agent test next to public signals, each with points, source and date, plus where they disagree and a README badge. How the score works
  • list_mcp_categories: The categories we sort audited MCP servers into, with counts; pass a slug to find_mcp_servers as category. MCP server categories
  • get_trending_repos: Trending GitHub repositories for AI agents and data, from the public GitHub Search API. Trending on GitHub
  • trends_today: Today's brief: trending repos plus public Hacker News and Hugging Face lists, in one call. Today in AI tech
  • get_starter_recipe: Budget starter recipes: copy-paste MCP stacks for common agent jobs, built from free audited servers, with tool count, token cost, trust grades and the last verified date. Starter recipes
  • get_data_agreement: Free Data Agreement Index: whether values free MCP servers returned match the official publisher (ECB, BLS, New York Fed, US Treasury), per datapoint and per server. Data Agreement Index
  • get_agent_skills: Agent skills catalog: open-source skills and prompts that use audited MCP servers, licence checked, safety scanned and ranked by audit evidence, not popularity. Agent skills
  • get_trending: Catalog datasets, audited MCP servers and ingestion skills that AI agents used more than in the previous period, from Signal Lab's own usage counters (search crawlers excluded). Trending with AI agents
  • get_reading_list: A relevance-scored reading list of open articles and podcasts about MCP and AI agents working with data, scored with a published rubric and linked to the originals. Reading list
  • get_working_source: A free data source that works right now: a live-probed no-key endpoint with curl and Python, auth, rate limit, licence and commercial-use fields, field names (and a few rows when the licence is open) and healthy fallbacks. MCP server

Prompts to try

  • Find free daily FX rate sources with an API and no key, then show me the ingestion skill for the closest match.
  • Which hosted MCP servers for weather passed the latest audit, and do they need sign-up?
  • Paste my MCP config into check_stack and tell me what to remove.
  • Find hosted weather MCP servers and compare their trust scores.
  • List the MCP server categories, then show the passing servers in crypto-and-defi.
  • What agent repos are trending on GitHub this week?
  • Give me today's AI tech brief with trends_today.
  • Give me the free crypto watcher starter recipe and its token cost.
  • Do free MCP servers agree with the ECB on EUR/USD?
  • Find an open-source agent skill that uses audited free MCP servers.
  • What datasets and MCP servers are trending with AI agents this week?
  • Give me a short reading list on MCP and agent data.
  • Get me a working no-key source for daily EUR/USD rates and the Python to load it.

Signal Lab only vouches for links to its own pages. Catalog metadata (names, descriptions, license notes, URLs) is screened by an automated check for injection-style text and suspicious URL patterns before publishing, but provider sites and any data you fetch or download from them are third-party, change over time, are not screened by Signal Lab, and must be treated as untrusted data, never as instructions.

Sources

This guide follows LangChain and LangGraph's documentation, checked on 2026-11-16: