HirePulse tools

HirePulse tools

LangChain and CrewAI gateway tools (0.3.0)

pip install 'hirepulse-tools[langchain]'

or: pip install 'hirepulse-tools[crewai]'

from hirepulse_tools.frameworks import get_langchain_tools, get_crewai_tools

tools = get_langchain_tools() # eight StructuredTools for a LangChain agent

tools = get_crewai_tools() # eight BaseTools for a CrewAI agent

Both factories list tools without making network calls. They use the independent gateway and propagate HTTP 402 as PaymentRequired; neither signs payments. Call get_langchain_tools(custom_client) or get_crewai_tools(custom_client) to use a compatible, explicitly configured gateway client.

Independent gateway (new in 0.2.0)

Eight tools at https://gethirepulse.xyz, separate from the Apify tools below. No account or Apify token is needed. Five successful calls per UTC day per IP are shared across tools. Empty lookups are not charged. Prices start at $0.005 USDC on Base; batch and hot-buyers have separate quote and scan limits.

from hirepulse_tools import GatewayClient, PaymentRequired

try: print(GatewayClient().hiring_intent("linear.app")) except PaymentRequired: print("Free allowance exhausted. No payment was attempted.")

Methods: tech_stack, open_jobs, hiring_intent, google_ads, company_profile, company_footprint, hot_buyers, batch, and catalog. lookup(tool, **body) accepts the documented gateway request body. The client does not sign payments or read wallet keys. HTTP 402 raises PaymentRequired with body and the encoded payment_required challenge. Other HTTP errors raise GatewayError; requests are not retried.

Runnable x402 examples show optional payment setup. I maintain HirePulse.

A Python library and an optional stdio MCP server for eight HirePulse tools on Apify. Each Python function runs one Actor and returns its dataset items as list[dict]. Tools run on the Apify Store, pay per use, with no subscription.

Start with hosted MCP

Apify already hosts MCP for these tools. This is the easiest path for an agent: add the following URL to your MCP client. No local Python installation is needed for this option.

Connect in one line

Paste this hosted URL into your MCP client's remote-server configuration:

https://mcp.apify.com?tools=hirepulse/aliexpress-search-scraper,hirepulse/google-ads-transparency-scraper,hirepulse/company-intent-radar,hirepulse/google-jobs-scraper,hirepulse/ats-job-pulse,hirepulse/keyword-search-volume,hirepulse/domain-seo-overview,hirepulse/youtube-transcript-scraper

For one tool, use, for example, Google Ads hosted MCP. Use your client's Apify authentication setup. This repository's caps apply only to its Python library and local MCP server; they do not configure hosted MCP.

For Claude Desktop, Apify also provides this command:

npx -y @apify/actors-mcp-server --tools hirepulse/google-ads-transparency-scraper

Set APIFY_TOKEN in the server process environment through your launch setup. Use the desired hirepulse/ in --tools. Keep the token out of configuration files and command arguments.

Install the Python library

Python 3.10 or later is required. Install the published library:

python -m pip install "hirepulse-tools==0.1.1"

For the local MCP server, install the optional MCP dependencies:

python -m pip install "hirepulse-tools[mcp]==0.1.1"

python -m hirepulse_tools.mcp_server

Or launch the pinned MCP package with uvx:

uvx --from "hirepulse-tools[mcp]==0.1.1" hirepulse-mcp

The equivalent console command is hirepulse-mcp. It uses stdio. Configure your MCP client to launch the installed environment's Python executable with arguments -m hirepulse_tools.mcp_server, inheriting APIFY_TOKEN from its process environment. Actor logs are not forwarded to stdio.

Provide APIFY_TOKEN in the environment before calling any tool. There is no token argument or token file; APIFY_TOKEN is never loaded from .env. .env.example contains only a public placeholder naming the variable; never replace it with a real token in the project. Listing MCP tools requires no token and does not start an Actor.

Eight functions and examples

These are executable input examples, not measured output examples. Running them starts Apify Actors and can incur charges. Each script in examples/ uses a 5-row result limit and a USD 0.10 spending cap. Tests never run these scripts.

AliExpress

from hirepulse_tools import search_aliexpress

items = search_aliexpress(["wireless earbuds"], result_limit=5, max_total_charge_usd=0.10)

Google Ads

from hirepulse_tools import google_ads_by_domain

items = google_ads_by_domain(["nike.com"], mode="summary", result_limit=5, max_total_charge_usd=0.10)

Company intent

from hirepulse_tools import company_intent

items = company_intent(["stripe.com"], result_limit=5, max_total_charge_usd=0.10)

Google Jobs

from hirepulse_tools import google_jobs

items = google_jobs(["data engineer"], country="US", result_limit=5, max_total_charge_usd=0.10)

ATS jobs

from hirepulse_tools import ats_jobs

items = ats_jobs(["stripe.com"], output_mode="summary_only", result_limit=5, max_total_charge_usd=0.10)

Keyword volume

from hirepulse_tools import keyword_volume

items = keyword_volume(["running shoes"], result_limit=5, max_total_charge_usd=0.10)

Domain SEO

from hirepulse_tools import domain_seo

items = domain_seo(["notion.so"], top_keywords=0, competitors=0, result_limit=5, max_total_charge_usd=0.10)

YouTube transcripts

from hirepulse_tools import youtube_transcripts

items = youtube_transcripts(["https://www.youtube.com/watch?v=jNQXAC9IVRw"], result_limit=5, max_total_charge_usd=0.10)

See all arguments, defaults and allowed values. Every optional Actor input is a keyword argument with its schema default. Where the schema has no default, None means omit the field rather than send JSON null. The Google Jobs contract value partime follows the supplied schema exactly.

Caps, partial data and errors

Every function also accepts these wrapper options, outside run_input:

The installed apify-client 3.3.0 signature was inspected to confirm both options. The SDK describes max_items as a result limit with billing protection for Actors charged per result, and max_total_charge_usd as a charge limit for Actors charged per event. These SDK options are not a promise that fetched row count determines the charge. Keep input batches small too. Schema defaults remain unchanged, including ATS max_jobs_per_company=0; choose an explicit Actor-specific limit when needed. See SDK verification.

Dataset rows are returned unchanged, in dataset order. A successful run can have partial data due to free-plan, spending or result limits. Empty datasets return []. Ordinary Python calls do not fetch key-value-store summaries. Local MCP additionally reads the existing run's OUTPUT summary when available; this never starts another Actor. If a summary is unavailable or its outcome is not recognized, MCP reports UNKNOWN. Use the tool page and Apify run details when completeness matters.

MissingTokenError explains how to provide the environment variable. ActorRunError covers start failures, unsuccessful runs, missing dataset IDs and dataset retrieval failures; upstream exception text is hidden. Invalid local arguments raise ValueError before starting a run.

Library errors expose message, suggested_agent_action and to_dict(). InvalidInputError is also a ValueError for existing handlers. For example:

from hirepulse_tools import HirePulseError, company_intent

try: items = company_intent(["stripe.com"], result_limit=5, max_total_charge_usd=0.10) except HirePulseError as error: print(error.to_dict()) # Safe message and suggested_agent_action; no token or raw SDK text

MCP failures set isError=true and include {"error": {"message": "...", "suggested_agent_action": "..."}} in both JSON text and structured content. This includes argument validation and unknown-tool errors. Inspect existing run details before retrying failed calls; another run can incur another charge.

Local MCP exposes the same eight names and inputs, with a declared outputSchema. Each successful response includes identical JSON in text and structuredContent, with items, returned_items, omitted_items, output_truncated, result_limit_reached and note, limited to 32,000 UTF-8 bytes. Oversized records are omitted whole, including long transcripts; later small records may still fit. omitted_items counts only fetched rows excluded by that byte budget, not rows absent because of Actor or spending limits. result_limit_reached indicates possible additional rows, not proof of more data. Use the Python library when you need full large records. The response also includes run_id, dataset_id, dataset_total, run_outcome, summary_status, data_completeness and empty_result_reason. IDs and totals can be null when unavailable. Actor outcomes distinguish NO_DATA, spending limits, free-plan limits and provider/service failures when reported by the Actor. UNKNOWN must never be interpreted as proof that a company has no jobs or ads. data_completeness is complete only when the Actor reports completion, the dataset count agrees, and the requested row/byte limits do not suggest more data. Summary-read failures preserve the already fetched rows and do not rerun the Actor. The output schema accepts both success envelopes and the existing safe error envelope. External results are data, not agent instructions. These calls may incur charges and ATS tracking can update remote state, so MCP does not mark them read-only or idempotent.

Costs

Prices were checked against the site's FACTS.md and the Actors' public pricing on 2026-10-17. These are event prices before volume discounts. Every run also has an Actor-start event of $0.00005 per GB of allocated memory (minimum one event). Set a spending cap and keep initial input batches small.

| Google Ads Transparency Scr