Ein MCP-Server für die Jobsuche unter einer Visa-Beschränkung.
Wenn du Sponsoring brauchst, beantwortet die meiste Job-Such-Software die falsche Frage. Sie sagt dir, dass eine Stelle existiert. Sie sagt dir nicht, ob der Arbeitgeber dich legal einstellen kann, ob der Titel „mid-level“ eigentlich eine Senior-Rolle verbirgt oder ob das Unternehmen, das die Stelle ausschreibt, eine Personalvermittlung ist, die überhaupt keine Lizenz besitzt.
Dieser Server beantwortet diese Fragen. Er stellt vier Tools über das Model Context Protocol bereit, damit ein KI-Assistent sie selbst prüfen kann, statt zu raten.
Tool | Antwort |
| Hält dieser Arbeitgeber eine UK-Skilled-Worker-Lizenz – und wie sehr kann ich dem Treffer vertrauen? |
| Gibt es hier harte Ausschlusskriterien? Sagt der Titel die Wahrheit über die Seniorität? |
| Was ist gerade live, abzüglich der Agenturen? |
| Habe ich mich hier bereits beworben? |
Related MCP server: JobFindsMe
Der naheliegende Weg, eine Sponsoring-Lizenz zu prüfen, ist, im Register nach dem Unternehmensnamen zu suchen. So hat es angefangen, und es ist falsch, oft genug falsch, um gefährlich zu sein.
Unternehmensnamen kollidieren. Um Pimberly mit dem eingetragenen Namen Pimberly Software Development Limited abzugleichen, entfernt man Wörter wie „Software“ und „Limited“. Dieselbe Normalisierung macht aus Minerva Defence ein minerva, das dreizehn nicht zusammenhängende Unternehmen trifft: einen Einrichtungsbetrieb, eine Kreditagentur, einen Anbieter für betreutes Wohnen.
Deshalb gibt check_sponsor einen Vertrauensgrad zurück, keinen booleschen Wert:
high – der markanteste Teil des Namens wurde exakt, einmal.
verify – etwas wurde gefunden, aber es könnte Zufall sein. Bestätige die eingetragene Rechtsform des Unternehmens und schlage diesen Namen nach.
none – nichts wurde nicht. Das ist kein Beweis, dass sie nicht sponsern können. Viele zugelassene Arbeitgeber treten unter einem Namen auf, der nicht dem eingetragenen entspricht.
Reale cases, die das geprägt haben:
Suchbegriff | Treffer | Was tatsächlich vorlag |
Minerva Defence |
|
|
Prevail | Prevail Technology Limited (Poole) | Der Arbeitgeber ist |
MAGIC | MAGIC SOFTWARE SERVICES LTD | Richtige Antwort, falscher Beleg. Der Arbeitgeber ist |
TransPerfect | PERFECT DIGITAL LTD | Der Treffer erfolgte über das Wort „perfect“ |
eFinancialCareers | eFinancialCareers Ltd | Richtig und nutzlos: lizenziert, aber eine Jobbörse. Stellen unter diesem Namen gehören not ebeneiche Dritten |
screen_posting den Text liest statt des TitelsEine Anzeige mit dem Titel „Software Engineer“ begann mit „As a Senior Software Engineer you will“ und bat jemand zu „the finding team“ zu stoßene. Eine andere mit dem Titel „Software Engineer (Java Mid)“ verwies drei Absätze weiter unten auf eine Senior-Stufe. Wer fon Titel filternen, übersieht beide.
screen_posting erkennt darüber hinaus harte Ausschlusskriterien, die kein Bewerbungscoaching aushebeln kann:
You must be a UK citizen and have lived in the UK for the past 10 years.
You must already hold high-level UK security clearance.und denjenigen, der mit einem Sponsoring am wichtigsten ist – von einem Unternehmen, das eine Lizenz mit A-Rating besitzt:
We are unable to offer visa sponsorship for this role. Candidates who need visa
sponsorship now or will need it in the future will not be considered.Eine Lizenz bedeutet, dass ein Unternehmen sponsern kann. Es ist nicht gesagt, dass es sponsern wird.
search_roles einen Funnel zurückgibtBritische Jobplatze werden von Personalvermittlungsagenturen dominiert, und diese Agenturen sind nicht Inhaber der Sponsor-Lizenz für die ausgeschriebene Stelle. Das Herausfiltern der Absender ist essenziell und zu brutal: Eine typische Suche schrumpf von sieben Suchergebnissen auf eines.
Ein einziges Ergebnis ohne Erklärung das fühlt sich wie Fehler an. Deshalb gibt lies dashboard alle, alle für each Stellung entfernt wurden:
{
"totalFromSource": 7,
"funnel": { "fromSource": 7, "afterAgencyFilter": 2, "afterSalaryAndAge": 1 },
"notes": ["5 of 7 results were recruitment agencies or job boards. ..."]
}Nicht oder inhalt nicht veröffentlichtee Gehälter werden behalten, never herausgefiltert. Rückhalt um Gehälter ist kein Beweis für niedrige Bezahlung.
npm install && npm run buildLade das aktuelle Register herunter (regelmäßig aktualisiert, ca. 11 MB):
https://www.gov.uk/government/publications/register-of-licensed-sponsors-workers
Speichere es als data/register.csv oder verweise SPONSOR_REGISTER_PATH darauf.
{
"mcpServers": {
"job-search": {
"command": "node",
"args": ["/absolute/path/to/job-search-mcp/dist/index.js"],
"env": {
"SPONSOR_REGISTER_PATH": "/absolute/path/to/data/register.csv",
"REED_API_KEY_FILE": "/absolute/path/to/.reed-api-key"
}
}
}
}Variable | Erforderlich | Zweck |
| nein | Register-CSV. Standard |
| für | Ort einer Datei verschlange Schlüssels. Vorzuziehen: |
| alternativ | Den Schlüssel direkt. Einفach, aber kopiert den Schlüssel in der MCP-Konfiguration |
| nein | NDJSON-Bewerbungsverlauf |
In normaler Betrieb werden keine Zugangs daten aus repo-internen Datei gelesen, und die Registersdaten liegen in .gitignore.
Oder mit der Claude-Code-CLI:
claude mcp add job-search --scope user \
--env SPONSOR_REGISTER_PATH=/path/to/data/register.csv \
--env REED_API_KEY_FILE=/path/to/.reed-api-key \
-- node /path/to/job-search-mcp/dist/index.jsnpm test54 Tests. Jedes Fixture ist eine echte Herstellung oder ein echt Registereintrag, der einen frühere Version dieses Codes geschlagen hat. Die Reed-Tests mocken fetch, sodass die Suite offline läuft und kein API-Kontingent verbraucht.
Zwei Fehler, welche die Suite finden konnte, als sie geschrieben wurde:
£40,000-85,000 wurde als Pauschalbetrag £40,000 eingespreist da Prob, weil der zweite Betrag keine Währungssymbol enthält. An einer Gehaltsschwelle entspricht das dem Unterschied zwischen „erreicht“ und „nicht erreicht“.
Ein Agenturfilter, entlang \brecruit\b geschrieben, traf funktional weder Recruitment noch Consultancy. Und ca. fünfzig Agenten sind einfach durchgerüchert.
core() egal – macht gezielt Beschreibung Sprung: By removal-Trages Remove descriptors deswegen Abgleich Markenkern und Firmierung funktional it. But genau das führt zu false positives. Der Vertrauensgrad weist den Preis diesen Trade-off aus, statt ihn zu verstecken.
Absenz ist kein Gegenbeweis. none bringt das Einspre aside Hinweis mit. The teuerste Error in this domain is die Folgerung unwilliger Arbeitgeber ist der sebstständiger operator – „die Search“ Fehler: none.
In anderen Worten: Wer einen nicht gelisteten Arbeitgeber nicht sponserungsfähig erklärt, macht den größten Fehler nei.
Read-only. Der Server liest das Bewerbungen-New-Legion, schreibt aber nicht. A Buchführungsergebnis ist eine Entscheidung, die ein Mensch muss.
MIT
Read the local application ledger. Use it before applying to avoid duplicate submissions to the same employer, and to see what has already been tried. Read-only.
| Name | Required | Description | Default |
|---|---|---|---|
| company | No | Filter to one employer; omit for the full history |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and aptly declares 'Read-only', telling the agent this is a safe, non-mutating operation. It also notes the data is a 'local' ledger. It doesn't specify edge cases (e.g., empty history) or return format, but for a simple read this is solid disclosure.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two tight sentences, front-loaded with the core purpose, then usage guidance and a behavioral flag. Every sentence earns its place with zero filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Complete for a low-complexity tool (one optional param, no output schema). It covers purpose, use case, and read-only behavior. The only gap is that the return contents aren't described, but for this simple ledger-read it's mostly sufficient.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the single optional parameter is already fully documented ('Filter to one employer; omit for the full history'). The description adds no parameter-level detail beyond this, which lands at the baseline 3.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource ('Read the local application ledger') that clearly distinguishes it from siblings like search_roles (searching postings) and check_sponsor (visa status). The purpose is immediately actionable and unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says when to use it ('Use it before applying to avoid duplicate submissions to the same employer, and to see what has already been tried'). It gives clear context, though it doesn't name sibling tools as explicit when-not-to-use alternatives.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Look up whether a company holds a UK Skilled Worker sponsor licence. Returns a confidence grade, because company names collide: a match on a generic word like 'global' or 'minerva' is not evidence. A result of 'none' does NOT mean the employer cannot sponsor; it means they are not listed under that name, and the registered legal entity should be checked instead.
| Name | Required | Description | Default |
|---|---|---|---|
| company | Yes | Employer name as it appears on the job posting | |
| skilledWorkerOnly | No | Restrict to the Skilled Worker route (default true) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden. It discloses that results include a confidence grade, that generic-word matches are unreliable, and that a 'none' result does not rule out sponsorship and should prompt checking the registered legal entity. This is substantial behavioral context beyond a basic lookup, though it does not cover aspects like rate limits or authentication.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three sentences and immediately states the action, then adds caveats about confidence and interpretation. Every sentence adds value, and the most important information (the lookup) comes first. No redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with no output schema and no annotations, the description covers the essential aspects: what it checks, what result to expect (confidence grade), and how to interpret negative results. It does not specify the exact response format, but the description gives enough for an agent to decide whether to call the tool and how to interpret its output. Some detail on the confidence grade scale might be missing, but overall it is quite complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so both parameters (company and skilledWorkerOnly) are already documented in the schema. The description adds no additional parameter-level detail; it only implies that company names may collide but does not clarify formatting or validation beyond what the schema states.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific action: 'Look up whether a company holds a UK Skilled Worker sponsor licence.' It names the resource (company) and the licence type, and the wording clearly distinguishes it from the unrelated siblings like screen_posting or application_history.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description does not explicitly specify when to use this tool versus alternatives, nor does it state exclusions. The purpose implies the use case (checking sponsorship), but there is no direct guidance on when not to use it or what to use instead. However, the description does provide interpretive guidance for results, which indirectly helps the agent decide next steps.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Screen a posting for hard eligibility bars (citizenship, residency, existing security clearance, explicit refusal to sponsor), for seniority the body implies but the title hides, and for salary and stack. Run this before investing effort in an application.
| Name | Required | Description | Default |
|---|---|---|---|
| title | Yes | Job title | |
| description | Yes | Full posting text: description plus requirements |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden of behavioral disclosure. It does disclose what the tool evaluates (the explicit criteria list), which is genuinely useful context. But it never states whether the call is read-only, what it returns, or any side effects — for an analysis tool with zero annotation coverage these gaps are felt.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a tight two sentences with no filler and front-loads the core purpose and criteria. The usage directive is placed at the end, which is acceptable given the value of the criteria list, though it could arguably lead with the 'run before applying' guidance for maximum salience.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple two-parameter analysis tool with no output schema, the inputs are fully covered and the screening categories are spelled out. However, the description does not indicate what the result conveys — whether it returns a verdict, a risk score, or flagged bars — leaving an agent unsure what to expect from the call.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the schema already documents both title and description fully; this earns the baseline 3. The description adds marginal semantic value by signaling why the parameters matter (the analysis centers on the posting text), but does not add syntax or format detail beyond what the schema provides.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb ('Screen') and resource ('a posting') with enumerated screening dimensions: hard eligibility bars (citizenship, residency, clearance, refusal to sponsor), hidden seniority, salary, and stack. This precise scope clearly differentiates it from siblings like check_sponsor, which covers only the sponsorship angle.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
'Run this before investing effort in an application' gives explicit timing and tells the agent to call it early in the workflow. However, it does not name alternatives or state exclusions — e.g., it never routes sponsorship-only cases to check_sponsor — so the guidance stops short of full alternative differentiation.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search live UK job listings via Reed, excluding recruitment agencies and job boards, which cannot sponsor roles they merely advertise. Returns a funnel showing what each filter removed, so a small result set can be explained rather than mistaken for an error. Requires REED_API_KEY.
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No | Maximum listings to return (default 25) | |
| keywords | Yes | Search terms, e.g. "react developer" | |
| location | No | UK town or city; omit for nationwide | |
| maxAgeDays | No | Maximum posting age in days (default 30) | |
| minSalaryGBP | No | Drop roles whose published band tops out below this. Unpublished salaries are kept. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description must disclose behavioral traits. It does: it states the auth requirement (REED_API_KEY), explains the funnel output (showing what each filter removed so a small result set is understandable), and notes the exclusion logic. These go beyond the schema and help the agent anticipate how the tool behaves. It lacks details on error handling or rate limits, but core behaviors are covered.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences. The first sentence states the core purpose and exclusions; the second explains the funnel and the auth requirement. No wasted words. The information is front-loaded: an agent gets the gist immediately, then the important behavioral details. This is exemplary conciseness.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given there is no output schema, the description explains the return format (a funnel) and the API key requirement. The parameters are fully documented elsewhere. The only minor gap is the precise structure of the funnel output, but that is not essential for calling the tool correctly. The description covers the key operational aspects an agent needs.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
All five parameters are fully described in the input schema (coverage 100%), so the baseline is 3. The description does not add parameter-specific guidance beyond what the schema provides; it mentions the funnel generally but doesn't tie specific parameters to outcomes. Since the schema already handles semantics, there is no deficit requiring compensation.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description opens with a specific verb-resource pair ('Search live UK job listings via Reed') and adds a key scoping detail: exclusions of recruitment agencies and job boards, with the reason they cannot sponsor. This distinguishes it from sibling tools (check_sponsor, screen_posting, application_history) whose purposes are plainly different. An agent can tell exactly what this tool does and for whom.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives context: it searches live UK roles on Reed and excludes entities that can't sponsor, implying it is used to find roles that can sponsor visas. It also explains the funnel output, which helps an agent interpret results. However, it doesn't explicitly state when to prefer this over siblings or when not to use it, though the tool names make alternatives obvious. The use case is well implied, if not formally stated.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0application_historycheck_sponsorscreen_postingsearch_rolesScored across 4 tools
Each tool addresses a clearly distinct aspect: sponsor verification, posting screening, application history, and job searching. No overlap in purpose and each has a specific trigger condition.
Three tools follow a verb_noun pattern (check_sponsor, screen_posting, search_roles), but application_history deviates as noun_noun. The naming is still descriptive and predictable, with only minor inconsistency.
Four tools is well-scoped for a focused job-search assistant, covering the core workflow without bloat. Each tool contributes a distinct function and none seem redundant.
The surface covers search, screening, sponsor checks, and local history, which addresses the primary workflow. A minor gap is the lack of a tool to update or add to the application ledger, but this may be handled externally.
Job search, application materials, and application tracking for AI assistants.
Job platform for AI agents. Track tech jobs from companies that match your stack.
Analyze job listings against your resume, track applications, and generate cover letters.
AI job search MCP — fact-checked jobs, application tracker, alerts. ChatGPT, Claude, Cursor.
