Visa sponsorship data your agents can trust.
Filings turns the Department of Labor’s public LCA and PERM files into one MCP server. Your agent sends a messy company name and gets back the right employer, certified wages and a citation. When a name is ambiguous, it gets candidates instead of a guess.
Keys are issued by hand during the pilot · Works with Claude, Claude Code, Cursor and any MCP client
- Employer
- Google LLC
- Cases
- 5,377 certified
- Period
- Jul 2025 – Jun 2026
Why Filings
Your agent’s answers, finally right.
H-1B and green card data that agents can trust, without the spreadsheet work.
Ask about any company. Get the right one.
Brand names, typos, old names, even an email address. Filings knows who you mean, and when it isn’t sure, it asks.
See the whole company, not just one entity.
Big employers file under many names. Filings shows each one’s own numbers and how they add up to the group.
From visa to green card.
Not just H-1B: see who sponsors employees all the way to permanent residence.
Spot who’s hiring more, and who stopped.
Filings since 2019, ready to compare. Find growing sponsors, new ones, and the ones that went quiet.
Answers, not spreadsheets.
Median pay, top employers, the share of new hires: Filings hands your agent the finished number, so it never has to add things up.
Straight about what it doesn’t know.
No made-up lottery odds or approval rates. Every answer says where it comes from and how recent it is.
90–95% fully correct · at most 3 wrong figures in 100 · about half the cost per correct answer, vs. an agent searching the web itself. See how we measured →
Benchmark
Fewer wrong answers than web search.
Same agent, two sets of 105 questions, November 2026, scored without the caveat rule (it favours Filings). The reference figures come from the DOL data Filings serves, which favours Filings. Setup, grading and limits →
Connect
One URL. Your agent is connected.
Filings is a hosted MCP server. There is nothing to install or download: add the URL and your API key to your client, and the tools appear.
Building your own agent? The same tools are on the REST API. Read the docs →
claude mcp add --transport http filings \ https://filings-api.fly.dev/mcp \ --header "Authorization: Bearer YOUR_KEY"
Why agents can rely on it
When a name is ambiguous, Filings doesn’t guess.
“Delta” is an airline, a software firm and two dental insurers. A spreadsheet lookup picks one and moves on. Filings returns assigned: false with the most likely candidates, so your agent asks the user instead of reporting the wrong company.
A wrong employer in a recruiting or sales workflow costs more than a follow-up question. That is the trade-off we make every time.
"assigned": false, "reason": "ambiguous", "candidates": [ // name, confidence "Delta Air Lines, Inc." 0.6 "DELTA SYSTEM & SOFTWARE, INC." 0.3 "Delta Dental Of California" 0.3 "Delta Dental Insurance Company" 0.3 "Delta Computer Consulting" 0.3 ]
The public file vs Filings
The data is public. Making it usable is the product.
| Dimension | Raw DOL disclosure file | What your agent gets from Filings |
|---|---|---|
| Employer names | The same company under different spellings and entities, quarter after quarter | One stable employer_id, with a confidence and the reason it matched |
| Ambiguous names | Nothing tells you that “Delta” is several different filers | No assignment, ranked candidates, so the agent can ask |
| Corporate groups | Legal entities are separate rows; there is no group total | Entities roll up into groups for rankings, with a warning whenever an answer covers one entity only |
| Wages | One row per case, in hourly, weekly or yearly units | Certified p25, p50 and p75, withheld when there are fewer than 5 cases |
| Freshness and source | Large Excel files to download from dol.gov every quarter | as_of, the data window and the OFLC citation on every answer |
Methodology
How we resolve an employer
Three rules, applied to every lookup. The reason for each decision comes back in the answer, so your agent can show its work.
Match on name, alias and state
We normalise the name and compare it with legal names and a maintained alias list, using the state when the agent passes it. No language model runs when your agent asks. The alias and group lists are reviewed with the help of AI models, and a name is added only with documented evidence. Every match returns a confidence and a reason.
Assign only when it’s certain
Only a unique match with a confidence of 0.80 or higher gets an employer_id. Anything else comes back unassigned, with candidates and the reason for each.
Keep entities and groups apart
Lookalike companies stay apart: a shared brand is not a reason to merge. Entities roll up into their group for rankings, and every entity-level answer says it is not the group total.
Kept apart, on purpose
- Amazon.com Services≠AWS≠Amazon.com Inc.
- Waymo LLC≠Google LLC
- Deloitte Consulting≠Deloitte Tax≠Deloitte & Touche
How agents use it
Three calls. Then stop.
Agents resolve the employer first, then fetch only what they need. Resolving first means a wrong guess never reaches a search.
- 01Is this the same employer?
identify_employerwith the name and state. If it abstains, the agent picks from the candidates instead of guessing. - 02What did they certify?
search_casesby employer_id: titles, wages and worksites, certified cases only. - 03What is this case?
get_caseby case number: one case, its as_of date and the resolved employer.
Need pay ranges? Add wage_summary. It only returns percentiles when there are at least 5 certified wages.
identify_employer(query="google", state="CA") # → emp_us_google_llc · 0.96 · alias+state search_cases(employer_id="emp_us_google_llc", job_title="Software Engineer") # → certified cases: title, wage, worksite wage_summary(employer_id="emp_us_google_llc", worksite_state="CA") # → p25 / p50 / p75, only if n ≥ 5 get_case(case_number="I-200-25267-331421") # → one case, as_of, resolved employer
Tools
Twelve tools your agent picks from
A messy company name, plus state if known, becomes an employer_id. Ambiguous names return candidates.
LCA and PERM totals for one employer, with sample cases.
Certified cases by employer, title, city, state, visa class or wage level.
One LCA or PERM case in full, by case number.
Certified wage percentiles. Withheld below 5 cases.
Rankings over all matching cases: top sponsors, highest median wage, by city or title.
The most-filed job titles for an employer, a place or a visa class.
Every DOL release that is loaded, with as_of dates and row counts.
The official OFLC source, publication date and caveat for the current window.
Does a company sponsor, for a role and in a place? A finished answer with certified counts, the latest decision and how the name was matched.
Where a base salary falls among certified wages for a role and place, with the prevailing wage per level.
Change over fiscal years: per-year counts, growing and declining sponsors, new and stopped ones.
Ask it like you would ask a colleague
Which companies in Austin filed LCAs for data engineers last year, and at what pay?
Has this employer filed PERM cases for this role before?
Resolve the 400 company names in this CRM export and flag the active sponsors.
Top 10 sponsors of registered nurses in Texas.
Median certified wage for level II software engineers in Seattle.
How did this group’s filings change between FY2024 and FY2025?
Coverage
Seven fiscal years, refreshed every quarter
The disclosure files of the Office of Foreign Labor Certification from October 2019 on. When DOL publishes a new quarter, we load it, re-run the matching and test it before it goes live. Your agent can check what is loaded with list_releases.
Pricing
Priced per tool call, not per seat.
Prices are not final and nothing is on sale yet. Keys are issued by hand during the pilot, and no payment is taken.
Scope
What Filings is, and what it isn’t
Independent
Filings is not affiliated with the U.S. Department of Labor. We republish its public disclosure files.
Determined cases only
A certified LCA is not a USCIS H-1B approval. Pending cases are not in the files.
Data, not advice
Filings tells you what was filed and certified. It does not give legal or immigration advice.
Cited, every time
Each answer names its data window, as_of date and the OFLC release it came from.
Give your agent the employer, not the spreadsheet.
Keys are issued by hand during the pilot. Connecting takes one command.