Skip to content

Spanish Top Tech Companies MCP

Every salary negotiation starts with a number the person on the other side has and you do not. A candidate in Spain asking for 62,000 needs to know that the median in the market is 55,000 and that the middle half runs from 42,000 to 72,000, because that is what the employer's recruiter is looking at. A hiring manager approving a budget needs to know that the same title pays 126,000 median at Shopify and 105,000 at Meta.

Spanish Top Tech Companies is a free, open-source project that answers both questions from Levels.fyi submissions and publishes the answer as an MCP server. It is keyless, and it is honest in a way that is unusual for this class: the medians are reported only where five or more people back them, fewer than five is marked, and the tool says so before you quote a number.

Server type:   remote Streamable HTTP, JSON-RPC 2.0, session header issued
Endpoint:      https://spanishtoptechcompanies.com/mcp
Server:        "tech-pay-spain" titled "Spanish Top Tech Companies", 1.20.1,
               protocol 2025-06-18, capabilities: tools
Registry:      io.github.pugarte7/spanish-top-tech-companies 1.20.1, active,
               published 2026-11-14
Auth:          none
Tools:         8
Currency:      euro, gross base per year; totals add bonus and stock
Rounding:      to the nearest 1,000
Source:        Levels.fyi submissions for people working in Spain

The evidence tier

Keyless, session-issuing, and every tool that answers for a person's question was driven. Seven tools/call requests were made across five tools; the two write-adjacent tools (report_application, application_guide) return instructions rather than touching an account.

What a role pays across Spain

salary_stats for software-engineer, the widest selection the tool takes without a company filter:

Measure Value
People behind the figure 4,925
Median base 55,000 EUR
25th percentile 42,000 EUR
75th percentile 72,000 EUR
Median total (base plus bonus and stock) 60,000 EUR
Share at or above 60,000 base 45.9%

The response carries its own caveat, and it is the kind that saves an argument: Backend is counted as software-engineer on Levels.fyi, so these figures are for every software engineer, not only Backend.

list_filters returns the vocabulary the other tools accept: 35 roles and a region map keyed on the Spanish autonomous-community codes (CT Catalonia, MD Madrid, AN Andalusia and the rest), so a narrowing of the selection is a filter rather than a free-text guess.

What a single company pays, by seniority

company_salaries for Glovo returned 441 salaries in Spain across Barcelona and Madrid, median base 70,000, p25 57,000, p75 80,000, median total 77,000, and 72.8% of that population at or above 60,000 base. The same call breaks the population into year bands:

Years People Median base Median total
0 to 1 17 35,000 35,000
2 to 4 91 53,000 57,000
5 to 7 165 68,000 75,000
8 to 11 111 78,000 87,000
12 to 19 55 88,000 100,000
20+ 2 94,000 115,000 (few)

The bands add up to the total: 17 + 91 + 165 + 111 + 55 + 2 = 441, the same figure the summary line reports. That is the arithmetic control this guide ran by hand, and it held, which is more than most compensation tools survive.

top_companies ranks 643 companies by median base for one role, with two rules stated in the response: a company with fewer than three salaries in the selection sorts after every company with three or more, and few: true marks a figure from fewer than five people. The head of the list for software engineers:

Company People Median base Median total
Shopify 18 126,000 127,000
Automattic 3 (few) 120,000 120,000
Workato 12 120,000 120,000
Adobe 7 110,000 166,000
Meta 11 105,000 -

Adobe is the row worth reading twice: a 110,000 base against a 166,000 total, because that is what equity does to a Spanish offer.

The jobs half: from a posting to a filled form

find_jobs is a live index, not a cached table. On 2026-11-14 it reported 7,523 postings in the index; filtering to software-engineer gave 1,595 matched postings merged into 1,438 results, and the free-text query python gave 1,135 into 1,051. The grouping rule is stated in the response: one company's postings with the same title, ignoring the places it names, count as one result, and the variants ride in also_posted_as.

The chain past the posting is where this server is different from a job board:

  • job_questions reads the application form of one posting, live from its board, so the agent knows what the employer actually asks before it starts drafting.
  • application_guide returns the procedure for applying on someone's behalf rather than doing it silently.
  • report_application records the outcome after sending, or after failing to, so the record is part of the system rather than a note in a chat.

Two honesty markers are worth recording because they are policy, not copy. The index states salary_data_as_of: 2026-11-09, five days before the run, next to a generated_at of the same minute - the salary layer and the postings layer are refreshed on different clocks and the response says which is which. And the self-reported rule is enforced: a selection with fewer than five people returns no percentile at all rather than a confident-looking number built from two.

A note addressed to the person, not the agent

Every find_jobs response carries a note_for_the_person block. It is a message from the maintainer to the human in the conversation, asking for a GitHub star, and its own instructions to the agent are explicit: pass it on once, under the postings, never star without the person's yes, and if a state document is kept, write the date and their answer on the Support line so it is not repeated.

That is the cleanest version of this pattern in the catalog. It solves a real problem - a free tool maintained by one person needs distribution - without turning the agent into a marketing channel, because the request is scoped, addressed to the wrong party on purpose, and carries its own do-not-repeat rule.

What does not work

  • company_salaries rejects company and requires name. The error is a readable sentence (name is required: a company such as Glovo) rather than a schema dump, which is the right way to fail but still a name an agent will guess wrong first.
  • company_salaries with no company in the table returns an error, not an empty set.
  • Spain only. There is no fallback country, and region takes Spanish community codes rather than free text.
  • Salary figures are rounded to the nearest 1,000 and come from self-submitted Levels.fyi data, so they are reported medians rather than payroll records.
  • The salary layer lags the postings layer by five days in this measurement, and the response says so rather than blending them.

Frequently asked questions

Does this server require an API key or an account?

No. initialize and all seven tools/call requests in this pass answered without a credential, and the server issues a session header rather than a token.

Are these figures reliable enough to quote in an offer?

They are medians from self-submitted data, rounded to the nearest 1,000, with a stated minimum of five people behind any percentile. Quote them as a market range with the sample size attached, which is how the tool presents them.

Can it apply for a job on my behalf?

It can read the form, prepare the application and record the outcome. The guide tool sets out the procedure for applying on someone else's behalf and the server treats the send as a step the person approves.

What is the difference between median_base and median_total?

median_base is gross salary. median_total adds bonus and stock, which is why a company like Adobe shows 110,000 against 166,000 and a cash-heavy company shows the two figures nearly equal.

Limits recorded

Spanish market only, euro only, Levels.fyi sourced. The salary layer and the postings layer have separate refresh clocks and the response reports both. Coverage of a company depends on how many people at that company submitted, so a company with three salaries is marked few and should not be compared against one with eighteen.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing