LangChain & CrewAI Job Search Tool for Real-Time ATS Data
Give LangChain, LangGraph, and CrewAI agents deterministic tool-calling access to 434 employer ATS platforms—returning structured USD salaries, hiring departments, extracted skills, and direct apply URLs without brittle HTML scraping.
Runnable langchain-core @tool Definition (Python 3.10+)
@tool("search_open_ats_jobs") with Typed Pydantic Schema
Install minimal dependencies (pip install langchain-core pydantic requests) and set JOBPOSTINGS_API_KEY (get 1,500 instant Free Credits). Want zero-code Streamable HTTP MCP instead? Jump to /mcp (9 auto-discovered tools) →
import os
from typing import Optional
import requests
from pydantic import BaseModel, Field
from langchain_core.tools import tool
JOBPOSTINGS_API_KEY = os.environ["JOBPOSTINGS_API_KEY"]
API_BASE = "https://api.jobpostingsapi.com/v1"
class AtsJobSearchInput(BaseModel):
q: Optional[str] = Field(
default=None,
description="Role title or boolean keyword query, e.g. 'staff backend engineer' or 'kubernetes'",
)
company: Optional[str] = Field(
default=None,
description="Company slug or corporate domain, e.g. 'stripe' or 'anthropic.com'",
)
country: Optional[str] = Field(
default="US",
description="ISO-2 country code filter such as 'US', 'GB', 'CA', or 'DE'",
)
remote: Optional[bool] = Field(
default=None,
description="Set True to restrict to remote-eligible ATS requisitions",
)
salary_min: Optional[int] = Field(
default=None,
description="Minimum annualized USD compensation floor, e.g. 160000",
)
limit: int = Field(
default=5,
ge=1,
le=25,
description="Maximum open ATS job postings to return (1 Credit per returned job; 0 Credits if empty)",
)
@tool("search_open_ats_jobs", args_schema=AtsJobSearchInput)
def search_open_ats_jobs(
q: Optional[str] = None,
company: Optional[str] = None,
country: Optional[str] = "US",
remote: Optional[bool] = None,
salary_min: Optional[int] = None,
limit: int = 5,
) -> list[dict]:
"""Search live open job postings across 434 ATS & corporate career portals (Greenhouse, Lever, Ashby, Workday, iCIMS).
Returns normalized titles, company domain, hiring department, annual USD salary bounds, skills, and direct apply_url."""
params: dict[str, str | int] = {"limit": limit, "include_description": "false"}
if q:
params["q"] = q
if company:
params["company"] = company
if country:
params["country"] = country
if remote is not None:
params["remote"] = str(remote).lower()
if salary_min is not None:
params["salary_min"] = salary_min
resp = requests.get(
f"{API_BASE}/jobs/search",
headers={"Authorization": f"Bearer {JOBPOSTINGS_API_KEY}"},
params=params,
timeout=15,
)
resp.raise_for_status()
rows = resp.json().get("data", [])
return [
{
"id": j["id"],
"title": j["title"],
"company": j.get("company", {}).get("name"),
"company_domain": j.get("company", {}).get("domain"),
"is_staffing_agency": j.get("company", {}).get("is_staffing_agency"),
"department": j.get("department"),
"work_arrangement": j.get("work_arrangement"),
"location": j.get("location", {}).get("raw"),
"salary_usd": (
f"${j['salary']['annual_usd_min']:,}–${j['salary']['annual_usd_max']:,}"
if j.get("salary") and j["salary"].get("annual_usd_min")
else None
),
"skills": (j.get("skills") or [])[:8],
"ats_source": j.get("source", {}).get("name"),
"apply_url": j.get("apply_url") or j.get("url"),
"posted_at": j.get("posted_at"),
}
for j in rows
]
Engineering Highlights & Credit Safeguards
- →Typed Pydantic tool schema: AtsJobSearchInput constrains LLM tool calls to validated q, company, country, remote, salary_min, and limit arguments.
- →Context-window safe by default: passes include_description=false during discovery searches so 5–25 structured job summaries fit in <1,200 tokens, reserving GET /v1/jobs/{id} for deep JD inspection.
- →Works identically across LangChain, LangGraph create_react_agent, and CrewAI tool lists—or zero-code via MultiServerMCPClient pointing to /mcp.
- →1,500 instant Free Credits ($1.50, no card) + $2.00 Sandbox Streak bonus let you prototype multi-step recruiting and GTM research agents immediately.
1. Choosing Between Custom @tool (Python) vs. MultiServerMCPClient
Both integration patterns hit the same sub-100ms indexed gateway; pick the pattern that matches how your agent orchestrates tool schemas and token budgets:
| Architecture Criteria | Custom @tool (requests + Pydantic) | MultiServerMCPClient (/mcp) | Headless Scraper Actors |
|---|---|---|---|
| Token Footprint per Result | ~120 tokens/job (curated dict projection) | Full gateway JSON (~350 tokens/job across 9 tools) | 3,000–12,000 tokens of raw noisy HTML |
| Exposed Endpoints | Tailored to /v1/jobs/search (extendable) | All 9 tools auto-discovered (search, job-count, feeds, ATS detect) | Single brittle site-specific scraper |
| Median Tool Call Latency | <100 ms cached / ~358 ms 30d p50 | <120 ms Streamable HTTP JSON-RPC | 15–90 seconds container cold start |
| Unit Pricing | From $0.20/1K jobs ($0.0002~$0.00076/job · $0 empty) | Identical credit ledger ($0 on discovery handshakes) | $2–$10/1K + residential proxy fees |
2. Wiring into LangGraph / CrewAI or Loading All 9 Tools via /mcp
Bind search_open_ats_jobs directly into LangGraph's create_react_agent, or use langchain-mcp-adapters to mount our remote Streamable HTTP MCP server in 10 lines:
Option B — Minimal MultiServerMCPClient Adapter (See /mcp for Full 9-Tool Spec)
import os
from langchain_mcp_adapters.client import MultiServerMCPClient
async def load_jobpostings_mcp_tools():
client = MultiServerMCPClient({
"jobpostingsapi": {
"transport": "streamable_http",
"url": "https://api.jobpostingsapi.com/mcp",
"headers": {"Authorization": f"Bearer {os.environ['JOBPOSTINGS_API_KEY']}"},
}
})
return await client.get_tools() # Exposes jobs_search, get_company_job_count, detect_company_ats, etc.
# Use the exact same @tool inside LangGraph or CrewAI agents:
from langgraph.prebuilt import create_react_agent
agent = create_react_agent(
model="anthropic:claude-sonnet-4-5",
tools=[search_open_ats_jobs],
)
result = agent.invoke({
"messages": [(
"user",
"Check whether stripe.com is hiring remote Data or Security engineers and list direct apply URLs."
)]
})
print(result["messages"][-1].content)
- Two-Stage Token Optimization for Recruiting Copilots: Have your agent call search_open_ats_jobs(include_description=False) to shortlist top matching job IDs, then call GET /v1/jobs/{id} only on the 1–3 finalist roles that require full description analysis.
- Add Sub-15ms Headcount Velocity Checks: Equip GTM research agents with GET /v1/companies/{slug}/job-count (or the get_company_job_count MCP tool on /mcp) to retrieve open_jobs, department breakdowns, and 30-day hiring delta for 1 Credit.
- Full Claude Desktop, Cursor & OAuth 2.1 Setup: For desktop IDE assistants, OAuth 2.1 PKCE connector URLs, and all 9 MCP tool schemas, visit our dedicated MCP Server Integration Hub (/mcp).
Frequently Asked Questions
- What Python packages are required to run search_open_ats_jobs?
- Only langchain-core, pydantic, and requests (`pip install langchain-core pydantic requests`). No custom SDK, browser binary, or proxy configuration is required.
- Can CrewAI agents use this @tool function directly?
- Yes. CrewAI natively accepts LangChain StructuredTool / @tool objects, or you can swap the decorator to from crewai.tools import tool with the exact same AtsJobSearchInput schema and HTTP request body.
- Where do I find the full 9-tool MCP specification and OAuth 2.1 connector?
- Visit https://jobpostingsapi.com/mcp for full MCP server documentation, Claude Code CLI commands, and OAuth 2.1 Dynamic Client Registration details.
Related Endpoints & Workflow Connectors
Other Connector · Clay HTTP API
Clay Hiring Intent Signals: Enrich Open ATS Jobs by Domain
From $0.20/1K ($0.0002~$0.00076/call) · 0 Clay credits · $0 on empty →
Other Connector · n8n Workflow JSON
n8n ATS Job Feed Workflow: Greenhouse & Workday to Slack
1 Credit / new event ($0.20–$0.76/1K) · 0 Credits when idle →
Instant API Key · Zero Credit Card Required
Test this LangChain & CrewAI connector free with 1,500 Credits ($1.50).
Includes 60 RPM / 5 RPS throughput, +2,000 Sandbox Streak bonus credits ($3.50 total free testing), and automatic 0-credit protection whenever a search returns 0 jobs.