Skip to content
Official MCP RegistryListed

Headcount Labs

Part ofHeadcount Labslisted on 3 directories

Job postings search: find, count and read open jobs from employer career sites, with pay and links.

First seen 15 Oct 2026. Evidence as of 16 Nov 2026.

9
Tools
From an anonymous probe
1
Source listings
Each with its own history
50
Recorded changes
Since first seen

Tools

ToolDescriptionBehaviour
count_jobsCount matching jobs without listing them. This is the live count; prefer it over market_lookup and market_breakdown for "how many" and "which employers are hiring". A count by title_advanced with no company, company_id or job_id is an estimate (count_is_estimate is true); if it is refused as too slow, add one of those three filters or count by title. count_jobs refuses these combinations unless company, company_id or job_id is also set: near together with title, title_advanced or q (use city with state instead; that counts fast); source, exclude_source, source_type, has_location, last_fetch_since, last_fetch_until or application_check_since together with title, title_advanced or q; has_description in any count. A refusal is invalid_request and is not temporary: change the filters as its message says instead of retrying the same call. Counting by description alone returns an estimate: count_is_estimate is true and count_margin gives the 95% range. search_jobs pages carry the same number as total when the filters are fast to count. Every error has code, message, retryable (true or false) and next_step; retryable errors also have retry_after_seconds. A call that runs past 25 seconds is stopped with query_too_broad: narrow it, do not repeat it. To answer "which companies are hiring for X" or "how are these jobs split by ...", pass group_by: employer (company works too), work_arrangement (remote, hybrid or in_person), title, state (US), city, country or employment_type. The answer then also has group_by, groups (the top 25 values by number of matching jobs, each with value and jobs), groups_from_jobs (how many matching jobs the breakdown is based on), groups_sampled (true when this is not an exact count of every match) and groups_unknown_jobs (jobs with no known value). The breakdown is exact up to 3,000 matching jobs, estimate from a random sample above. Either way the top 25 values are returned. Above 3,000, groups_are_estimates is true, groups_note states the sample size in plain words, and each jobs number is scaled from that random sample of matching jobs (picked by job id, which is random) up to the total, not a raw sample count. If the breakdown times out, the count is still returned with no groups and groups_note says the breakdown timed out. A job with several locations or postings can count under more than one value. group_by employer keeps its own keys: employers (each with employer_name, company_id and jobs), employers_from_jobs, employers_sampled and employers_unknown_jobs, and when the breakdown is an estimate, employers_are_estimates and employers_note. When a filter is too broad to count exactly in time, count is an estimate: the answer then has count_is_estimate: true, count_margin (the 95% range either side of count) and count_note.Read-only
get_jobGet a job by job_id. Default shape matches search_jobs, plus the full description, currentness, stale, last_checked_at, and linked postings. Pass view: "full" for every REST field, or fields to pick top-level properties (job_id is always included). One whole reply (every text part and structuredContent together) stays under about 50,000 characters. Each job lists its first 10 locations; locations_total and locations_truncated: true say when there are more. A search page that would not fit gets fewer jobs, with page_note saying so; next_cursor continues at the first job not shown, so no job is skipped and only jobs shown are counted. Only one job too big on its own is shortened (description_truncated; truncated.how_to_get_rest says where the rest is).Read-only
list_fieldsFind search_jobs filters and job fields matching keywords, with type and a short description. Pass words such as ["pay", "remote"]; no keywords lists everything. kind is filter, option (sort, a search_jobs argument beside filters) or field (postings[].x is a posting property). A field state object says whether the source published the value (known), Atlas derived it, the source omitted it, or extraction failed.Read-only
market_breakdownRank one dimension (by: title, state, city, company, seniority, employment_type or remote) by jobs, pay or another figure, optionally inside a filter. Daily snapshot (see as_of). For live counts or who is hiring now, use count_jobs. Cells with fewer than 30 jobs are omitted. No account needed.Read-only
market_dimensionsList dimensions and metrics for market_lookup, market_breakdown and salary_range; pass dimension (and optional q) to list that dimension's values. No account needed.Read-only
market_lookupLook up job count, pay, pay share, work arrangement, posting age, new postings or top employers for up to two of title, place, company, seniority, employment_type and remote. Omit metric for every figure. Daily snapshot (see as_of). For live counts or who is hiring now, use count_jobs. Cells with fewer than 30 jobs are omitted. No account needed.Read-only
salary_rangePosted pay percentiles (25/50/75/90) for a title, place (default US) and optional seniority, with sample size and a short recommendation. Daily snapshot (see as_of). For live counts or who is hiring now, use count_jobs. Cells with fewer than 30 jobs are omitted. No account needed.Read-only
search_docsSearch Headcount Labs docs (filters, limits, MCP setup, errors) and return matching sections with URLs. Works without an account.Read-only
search_jobsSearch current job postings by keyword, title, company, place, pay and date. Call list_fields for filter names. Country and employment_type accept comma-separated values or arrays. Pages return up to 50 jobs (limit, default 10), or 10 when descriptions is "full" (default 8, about as many whole descriptions as fit in one reply). Pass next_cursor as cursor for the next page. One search shows up to 250 jobs; narrow filters or use count_jobs for how many match. Default results are compact summaries (employer, title, locations, pay, dates, urls, snippet). Pass view: "full", fields, or get_job for more. descriptions: "snippet" (default) previews text; "full" returns whole descriptions. title matches one phrase in order. title_advanced and description_advanced are Boolean: an operator between every two words or groups, for example title_advanced: "(engineer OR developer) AND NOT manager" (write AND NOT, never a bare NOT after a word; keep at least one word that is not negated). sort: relevance needs q, title or description; with only title_advanced or description_advanced use updated. Each page also has total (how many jobs match, the same number count_jobs gives) and total_is_estimate (true when total is an estimate; total_margin is then the 95% range either side). total is null, with total_note saying why, when the filters are slow to count (description, description_advanced, title_advanced or q without company, near or source filters) or the count did not finish within the time the page took; then call count_jobs. One whole reply (every text part and structuredContent together) stays under about 50,000 characters. Each job lists its first 10 locations; locations_total and locations_truncated: true say when there are more. A search page that would not fit gets fewer jobs, with page_note saying so; next_cursor continues at the first job not shown, so no job is skipped and only jobs shown are counted. Only one job too big on its own is shortened (description_truncated; truncated.how_to_get_rest says where the rest is). Every error has code, message, retryable (true or false) and next_step; retryable errors also have retry_after_seconds. A call that runs past 25 seconds is stopped with query_too_broad: narrow it, do not repeat it.Read-only

Change history

  1. search_jobs: input schema changed
  2. search_jobs: description changed (+""full" (default 8, about as many whole descriptions as fit in one reply)." +"Each page also has total (how many jobs match, the same number count_jobs gives) and total_is_estimate (true when total is an estimate; total_margin is then the 95% range either side). total is null, with total_note saying why, when the filters are slow to count (description, description_advanced, title_advanced or q without company, near or source filters) or the count did not finish within the time the page took; then call count_jobs. One whole reply (every text part and structuredContent together) stays under about 50,000 characters. Each job lists its first 10 locations; locations_total and locations_truncated: true say when there are more. A search page that would not fit gets fewer jobs, with page_note saying so; next_cursor continues at the first job not shown, so no job is skipped and only jobs shown are counted. Only one job too big on its own is shortened (description_truncated; truncated.how_to_get_rest says where the rest is). Every error has code, message, retryable (true or false) and next_step; retryable errors also have retry_after_seconds. A call that runs past 25 seconds is stopped with query_too_broad: narrow it, do not repeat it." -""full".")
  3. get_job: description changed (+"One whole reply (every text part and structuredContent together) stays under about 50,000 characters. Each job lists its first 10 locations; locations_total and locations_truncated: true say when there are more. A search page that would not fit gets fewer jobs, with page_note saying so; next_cursor continues at the first job not shown, so no job is skipped and only jobs shown are counted. Only one job too big on its own is shortened (description_truncated; truncated.how_to_get_rest says where the rest is).")
  4. count_jobs: input schema changed
  5. count_jobs: description changed (+"search_jobs pages carry the same number as total when the filters are fast to count. Every error has code, message, retryable (true or false) and next_step; retryable errors also have retry_after_seconds. A call that runs past 25 seconds is stopped with query_too_broad: narrow it, do not repeat it.")
  6. count_jobs: input schema changed
  7. count_jobs: description changed (+"count_jobs refuses these combinations unless company, company_id or job_id is also set: near together with title, title_advanced or" +"(use city with state instead; that counts fast); source, exclude_source, source_type, has_location, last_fetch_since, last_fetch_until or application_check_since together with title, title_advanced" +"q; has_description in any count. A refusal is invalid_request and is not temporary: change the filters as its message says instead of retrying the same call.")
  8. server instructions changed (+"count_jobs also refuses near together with title, title_advanced or q, and source filters together with title, title_advanced or q, unless company, company_id or job_id is set: count around a place with city and state instead of near. A refused count is not temporary; change the filters, do not retry it.")
  9. search_jobs: input schema changed
  10. search_jobs: description changed (+"one" +"order." +"and description_advanced are Boolean: an operator between every two")
  11. list_fields: description changed (+"filter, option (sort, a search_jobs argument beside filters)" -"filter")
  12. count_jobs: input schema changed
  13. count_jobs: description changed (+"A count by title_advanced with no company, company_id or job_id is an estimate (count_is_estimate is true); if it is refused as too slow, add one of those three filters or count by title.")
  14. server instructions changed (+"How to query: 1. filters.title is one phrase in order; filters.title_advanced and filters.description_advanced are Boolean. 2. Boolean: an operator (AND, OR) between every two words or groups; write a AND NOT b, never a NOT b; keep at least one word that is not negated; put phrases in quotes. 3. sort: relevance needs q, title or description; with only title_advanced or description_advanced use updated. 4. count_jobs by title_advanced alone gives an estimate; if it is refused as too slow, add company, company_id or job_id, or count by title. 5. list_fields gives every filter name; search_docs gives the full syntax. Example: search_jobs {"filters":{"title_advanced":"(engineer OR developer) AND NOT manager","country":"US","work_arrangement":"remote"},"sort":"updated","limit":10}.")
  15. count_jobs: description changed (+"Counting by description alone returns an estimate: count_is_estimate is true and count_margin gives the 95% range." -"or description")
  16. search_jobs: description changed (+"pay" +"date." +"for")
  17. search_docs: description changed (+"Search" +"docs" +"return matching")
  18. salary_range: description changed (+"Posted pay percentiles (25/50/75/90)" +"with" +"short")
  19. market_lookup: description changed (+"Look up job" +"pay," +"share, work arrangement,")
  20. market_dimensions: description changed (+"dimensions and metrics for" +"salary_range; pass" +"optional")
  21. market_breakdown: description changed (+"optionally" +"filter. Daily" +"(see as_of). For live")
  22. list_fields: description changed (+"matching" +"(postings[].x" +"posting property).")
  23. get_job: description changed (+"Default" +"matches search_jobs, plus" +"full description,")
  24. count_jobs: description changed (+"matching" +"count; prefer" +"over market_lookup and market_breakdown")
  25. server instructions changed (+"Most-used tools: search_jobs (find postings), count_jobs (how many match, or which employers are hiring), get_job (one" +"by job_id). Then market_breakdown, market_lookup, list_fields, search_docs, market_dimensions" +"salary_range. These")
  26. search_jobs: input schema changed
  27. salary_range: input schema changed
  28. market_lookup: input schema changed
  29. market_dimensions: description changed (+"dimension,." -"dimension, such as the metro names.")
  30. market_breakdown: input schema changed
  31. market_breakdown: description changed (-"metro,")
  32. get_job: input schema changed
  33. count_jobs: input schema changed
  34. count_jobs: description changed (+"Counting by q or description text needs a title or company filter too, otherwise it is refused as too slow; add one and retry." +"city," -"city (metro works too),")
  35. server instructions changed (+"Bulk" +"full" +"are coming soon. Public subsets of it (ODbL 1.0) are at https://huggingface.co/HeadcountLabs.")
  36. salary_range: tool added
  37. market_lookup: tool added
  38. market_insights: tool removed
  39. market_dimensions: tool added
  40. market_breakdown: tool added
  41. count_jobs: description changed (+"This is the live count: it reads the job database as it is now, so use it for "how many jobs" and "which employers are hiring" questions, and prefer its numbers over market_lookup and market_breakdown, which read a daily snapshot.")
  42. server instructions changed (+"market_lookup, market_breakdown, salary_range, market_dimensions," +"pay ranges," +"share, posting age and rankings, market_lookup, market_breakdown and salary_range answer")
  43. search_jobs: description changed (-"(free)")
  44. search_docs: description changed (+"Works" -"Free, and works")
  45. market_insights: description changed (-"Free.")
  46. list_fields: description changed (-"Free.")
  47. server instructions changed (+"For bulk downloads of the whole job dataset (CC BY 4.0), see https://huggingface.co/datasets/HeadcountLabs/job-postings." +"filters" +"more.")
  48. search_jobs: input schema changed (+view)
  49. search_jobs: description changed (+"50" +"10), or up to 10 jobs when descriptions is "full"; a larger limit is refused. Pass" +"250")
  50. get_job: input schema changed (+view)
Source listings
SourceListingFirst seenLast seenVersions
Official MCP Registrycom.headcountlabs/jobs15 Oct 202616 Nov 20263