structuredlinkfetch
Fields an LLM can reason over
salary as numbers (min · max · currency · interval), applicant_count as an integer with a timestamped history, is_active as a bool. No string-soup, no parsing on your side.
live · api + mcp + dataset
§ data · jobs for ai
LinkedIn jobs for AI agents.
The jobs-data layer for the agent you're building. 45,000+ new roles indexed every day, delivered as a typed REST API, as MCP tools your agent calls natively, or as the whole 8M+-row archive in Postgres. Catch hiring signals without an enterprise sales cycle.
§ 01 · why ai builders choose this
mcp-nativelinkfetch
One config block in Claude or Cursor
linkfetch-mcp registers jobs search, job detail, applicant history, and location lookup as tools. Your agent searches and cites jobs without you writing a fetch wrapper.
flat-pricelinkfetch
Or take the whole archive
8M+ postings since April 2026 as a PostgreSQL dump, CSV, and JSONL for $199 one-time — or $49/month for live read-only Postgres. No credits, no meter.
§ 02 · three ways to plug in
REST. MCP. Postgres.
Pick the surface that matches your stack. Same rows, same provenance — different ergonomics. No SDK to install: the API is plain HTTPS and JSON.
curl · rest1 credit per call
# Senior+ machine-learning roles in the US posted this week
curl "https://api.linkfetch.io/v1/jobs" \
-H "Authorization: Bearer $LINKFETCH_KEY" \
-G \
--data-urlencode "q=machine learning" \
--data-urlencode "level=senior,staff,principal" \
--data-urlencode "location=United States" \
--data-urlencode "posted_within=week" \
--data-urlencode "limit=50"typescript · fetchground an llm in fresh roles
// Plain fetch — no SDK to install
const params = new URLSearchParams({
q: "machine learning",
level: "senior,staff,principal",
posted_within: "week",
limit: "50",
});
const res = await fetch(`https://api.linkfetch.io/v1/jobs?${params}`, {
headers: { Authorization: `Bearer ${process.env.LINKFETCH_KEY}` },
});
const { data: jobs, meta } = await res.json();
// Hand the typed rows to the model as grounded context
const summary = await anthropic.messages.create({
model: "claude-sonnet-5",
max_tokens: 1024,
messages: [{
role: "user",
content: `Summarize hiring trends from these ${jobs.length} roles (fetched ${meta.fetched_at}):\n${JSON.stringify(jobs)}`,
}],
});python · requestswatch a competitor
import os, requests
# Watch a competitor: everything they posted in the last 24h
r = requests.get(
"https://api.linkfetch.io/v1/jobs",
headers={"Authorization": f"Bearer {os.environ['LINKFETCH_KEY']}"},
params={"company": "acme-inc", "posted_within": "24h", "limit": 50},
)
jobs = r.json()["data"]
for job in jobs:
print(job["first_seen_at"], job["title"], job["applicant_count"])mcp · claude · cursor · zedno fetch wrapper, no parser
# claude_desktop_config.json
{
"mcpServers": {
"linkfetch": {
"command": "npx",
"args": ["-y", "linkfetch-mcp"],
"env": { "LINKFETCH_API_KEY": "sk_live_…" }
}
}
}
# Then ask Claude:
# "Find every infra-engineer role in Berlin posted this week.
# Group by company and rank by applicant velocity."§ 03 · the row
The shape your agent reasons over.
One JSON object per posting, deduped on linkedin_job_id. The same shape comes back from the API, the MCP tools, and the dataset. Field-level coverage is published on the dataset page: descriptions on every row, parsed salary on ~10%, applicant counts on ~8%.
get · /v1/jobs/:id1 credit · empty = free
{
"linkedin_job_id": "4072311928",
"title": "Staff Backend Engineer",
"company": "Vercel",
"company_id": 14080355,
"location": "San Francisco, California, United States",
"level": "Mid-Senior level",
"employment_type": "Full-time",
"salary_min": 220000,
"salary_max": 290000,
"salary_currency": "USD",
"salary_interval": "year",
"applicant_count": 117,
"applicant_count_is_capped": false,
"posted_at_precise": "2026-04-29T08:13:11Z",
"first_seen_at": "2026-04-29T08:14:18Z",
"is_active": true,
"canonical_url": "https://www.linkedin.com/jobs/view/4072311928/"
}§ 04 · vs the alternatives
Where the other vendors charge enterprise prices.
| capability | LinkFetch | Coresignal | Bright Data |
|---|---|---|---|
| 01Floor price | $0 · $5 free credit, then $29/mo | $49 / mo | $250 min order |
| 02Bulk export | $199 one-time · 8M+ rows | $1,000+ per dataset | $2.50 / 1K records |
| 03MCP server (Claude · Cursor · Zed) | Yes — linkfetch-mcp | — | — |
| 04Time to first call | 90 seconds, no card | Demo · eval · contract | Hours · proxy setup |
| 05is_active flag (re-checked on recrawl) | Yes | Yes | Not surfaced |
| 06Parsed salary fields | min · max · currency · interval | Yes | Range only |
| 07Applicant count + history | Yes — timestamped series | — | — |
| 08Provenance per row | source · fetched_at · freshness_days | Partial | — |
competitor figures from their public pricing pages · sources linked from each full teardown
§ 05 · agents you can ship this week
Three patterns, one data layer.
Each pattern is a recipe with a working prompt and tool chain. Same API key, same rows, different agent loop.
agent · 01recipe
AI recruiter
Ground a candidate-matching agent in fresh roles. Pull new postings into your vector store on a schedule; let the model reason over current demand, not last quarter's snapshot.
- ▸Poll posted_within=24h → embed → match
- ▸is_active flag keeps the index clean
- ▸Salary fields drive offer benchmarking
agent · 02recipe
Sales intent signal
Hiring is the loudest growth signal a company emits. Trigger your SDR agent when a target account posts a role that matches your buyer pattern.
- ▸Filter by company_id + role family
- ▸New posting → CRM enrichment → outbound
- ▸Applicant velocity says how urgent the hire is
agent · 03recipe
Market research
Watchlist of N companies, weekly LLM-written market brief. Hiring velocity, role-mix shifts, geo expansion — the alt-data signal, indie-priced.
- ▸Buy the dataset for the full archive
- ▸Long-form description for thematic analysis
- ▸Provenance on every row for citation
§ 06 · faq
What AI builders ask before they sign up.
What makes this AI-ready vs a generic jobs API?
Three things. Structured fields the model can reason over — parsed salary as numbers, level and employment_type as strings, is_active as a boolean, applicant_count as an integer — not a bag of strings. MCP-native, so Claude Desktop, Claude Code, Cursor, and Zed pick it up as a tool with one config block (linkfetch-mcp). And provenance on every row, so when your agent surfaces a job to a user it can cite where and when it learned it.
Why not just use Coresignal or Bright Data?
Coresignal is an enterprise data-warehouse product with a sales cycle; Bright Data has a $250 minimum order and doesn't surface an is_active flag. Both are built for procurement, not for an indie dev shipping an agent on a Tuesday. LinkFetch starts at $0 with a $5 credit, ships an MCP server, and sells the whole archive for $199 one-time if you'd rather have the rows in Postgres.
How fresh is the data?
We crawl LinkedIn's public guest endpoints continuously and index about 45,000+ new postings per day across 218 regions. Every row carries first_seen_at and last_seen_at, and is_active is re-checked on recrawl, so your index doesn't fill up with expired listings.
Do you have webhooks?
Not yet. Today you poll GET /v1/jobs with posted_within=24h (1 credit per call), or take the $49/month live database access and query first_seen_at directly. Real-time push delivery is on the roadmap; we'll announce it in the changelog when it ships rather than before.
Is this legal? Can I cite it from my agent?
Yes. We only index public job postings — listings any logged-out visitor can see. No fake accounts, no rented sessions, no cookie theft. We're aligned with GDPR and CCPA. Provenance fields (source, fetched_at, freshness_days) ship on every API row, so your agent can always say where it learned what it learned.
What does it cost to run an agent on top of this?
Flat per request, no per-result math. A jobs search is 1 credit and a job detail is 1 credit, whether the page holds one row or fifty. Empty results are free. The $5 free credit covers thousands of jobs calls; Starter is $29/month and Pro is $99/month. If you'd rather not meter at all, the dataset is $199 one-time or $49/month for live Postgres access.
Which LLM frameworks does this work with?
All of them — the REST API returns plain JSON, so any HTTP client works and there is no SDK to install. The MCP server (linkfetch-mcp) plugs straight into Claude Desktop, Claude Code, Cursor, Zed, and Continue. Most customers wire it as a tool inside their agent loop alongside web search and a vector store.
How does it compare to scraping LinkedIn jobs myself?
You can. LinkedIn's guest endpoints are public. But you'll spend a quarter building the parser, another on rate-limit and proxy management, and you'll still be missing the structured shape — parsed salary, level, industry_ids, applicant counts over time — that we extract per row. We sell the time, not the access.
ready when you are