Skip to main content
Glama

mcp-lever

npm CI license MCP Registry Glama M8ven LobeHub Install in Cursor Install in VS Code

Lever is recruiting software that thousands of companies use to run their hiring, and every customer gets a public job board that comes with it. Each board carries that company's open positions with their title, their location, the team and department they sit in, the commitment they ask for, the full advert, and the salary range where the company chose to publish one. Lever hosts one board per company, on either its global or its European instance, and publishes no index across them.

This server connects a chat client to those boards. You name the companies you are interested in, and it turns each name into the site name that addresses its board, searches their openings, filters them by location, team, workplace type, country, salary or how recently they were posted, reads one opening in full, and lists the wordings each company filters by. It needs no API key and no account.

Version française


Install

One-click install

Install in Cursor Install in VS Code

Claude Code

claude mcp add lever -- npx -y mcp-lever

Claude Desktop, Cursor, and any client using the standard config format

{
  "mcpServers": {
    "lever": {
      "command": "npx",
      "args": ["-y", "mcp-lever"]
    }
  }
}

Node 24 or later is required, and no environment variable has to be set.

With Docker

{
  "mcpServers": {
    "lever": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "ghcr.io/smeet666/mcp-lever:2.1.0"]
    }
  }
}

-i keeps stdin open, which is where the protocol travels, and -t is left out because a TTY rewrites the stream. The container needs outbound HTTPS to api.lever.co and api.eu.lever.co, and nothing else: no volume, no port, no credential.

Bundle, without npm

Download mcp-lever-2.1.0.mcpb from the latest release and open it. A client that supports MCP bundles installs it on its own, with no npm and no configuration file to edit. The bundle carries its dependencies, so nothing is fetched at install time.

Related MCP server: RecruitData

What you can ask

  • "Which of Included Health, Netlify and Ramp are hiring on Lever?"

  • "Find me remote engineering roles at those three companies."

  • "Read me that opening in full."

  • "What locations does Included Health list its jobs under?"

  • "Anything posted in the last two weeks at Netlify?"

Every question starts from a company, since Lever offers no search across boards. search_jobs resolves the names you give it, so no preparation is needed:

resolve_company(["Included Health"])  ->  includedhealth, global instance, publishing
search_jobs(["Included Health"], keyword: "therapist")
get_job("includedhealth", "6f97a19f-…")

Tools

Tool

What it does

resolve_company

Turns company names into the Lever site names of their boards.

search_jobs

Searches the openings of the companies you name.

get_job

Reads one opening in full, advert included.

list_filter_values

Lists the wordings one company files its openings under.

A Lever site name distinguishes case, so Flex answers where flex returns nothing. Four spellings are tried per name on each of the two instances, and the answer lists what was sent, so nothing found is never proof that a company is absent from Lever.

resolve_company

Turns company names into Lever site names, reporting every instance that answered. It takes a list.

Argument

Type

Required

What it does

names

array of 1 to 25 strings

yes

Company names, or Lever site names you already know.

In return: one entry per name, carrying input; found, a list of { slug, instance, publishes } where publishes is false for a site that exists and lists nothing today; tried, the spellings sent in order; and cached, true when this session had already resolved that name. A name answering on both instances comes back with both, and neither is elected: pass the one you mean to the other tools.

search_jobs

Searches the openings of the companies named. Lever applies the filters it supports on its own exact wording, and this server applies the rest to the openings it read.

Argument

Type

Required

What it does

companies

array of 1 to 25 strings

yes

Company names or Lever site names. Each is resolved here.

keyword

string

no

Words to look for in the title and the advert.

location

array of 1 to 20 strings

no

Locations, exactly as Lever writes them.

team

array of 1 to 20 strings

no

Teams, exactly as Lever writes them.

department

array of 1 to 20 strings

no

Departments, exactly as Lever writes them.

commitment

array of 1 to 20 strings

no

Commitments, exactly as Lever writes them.

workplace_type

array of 1 to 4 strings

no

remote, hybrid, onsite or unspecified.

country

array of 1 to 20 two-letter codes

no

Countries as ISO codes, as in FR or US.

salary_min

number, 0 or more

no

The lowest upper bound of a salary range to keep.

salary_interval

string

no

The period salary_min is written in, such as per-year-salary.

currency

three-letter code

no

The currency salary_min is written in, as in EUR.

posted_within_days

integer, 1 to 3650

no

How recent an opening must be.

limit

integer, 1 to 100, default 25

no

Openings to read per company.

skip

integer, 0 to 100000, default 0

no

Openings to step over per company.

Lever itself applies location, team, department and commitment; this server applies keyword, workplace_type, country, salary_min, salary_interval, currency and posted_within_days to what it read. list_filter_values publishes the wordings the first four take, and a wording Lever does not know comes back as an empty list.

In return: jobs, each carrying id and company_slug, which get_job takes, plus title, location, all_locations, country, workplace_type, team, posted_at, url and apply_url. commitment and department are absent when the company records neither. salary is null for an opening published without one, which is never the same as zero, and it carries the interval Lever wrote it in, never converted or annualised. per_company gives one outcome per company, with a status of read, unresolved, empty or failed, which are four different answers, and the read and returned counts around the filters. total_available is always null: Lever publishes no result count. The rows carry no advert text, since one company's board can run to megabytes.

limit applies per company, and a company whose openings fill it may publish more: the notes say when that happened, and that a count taken inside that window measures the window. posted_within_days walks up to five pages per company, and Lever pages by title, so an opening published yesterday can sit anywhere in a board.

get_job

Reads one opening in full: the advert, its named sections, and the salary as published.

Argument

Type

Required

What it does

company_slug

string

yes

The Lever site name, as resolve_company returns it.

job_id

string

yes

The identifier of one opening, as a search returns it.

instance

global or eu

no

The instance the row came from. The global one by default.

In return: job, holding the fields a search row carries, plus description, sections as { heading, items }, salary_note for what the company wrote beside the range, and source with the address it was retrieved from.

list_filter_values

Lists the team, location and commitment wordings one company uses. Read it before filtering: Lever matches its own wording, and the vocabulary belongs to each company, one writing Full-time where another writes EE Full-Time.

Argument

Type

Required

What it does

company_slug

string

yes

The Lever site name, as resolve_company returns it.

instance

global or eu

no

The instance this site lives on. The global one by default.

fields

array of 1 to 3 of team, location, commitment

no

Which vocabularies to read. Each costs one request, and all three are read by default.

In return: company_slug, instance, and fields holding a list of { value, count } for each vocabulary asked for. A count is null where Lever published no figure alongside the category.

Configuration

Nothing has to be configured. The server reads no environment variable, and the mcpServers block above is complete as written.

The pacing, the timeout and the cache are settings of the client layer, which As a library shows how to pass. The interval between two requests can be widened there and never narrowed.

Errors

Every failure carries one of six codes, a message, and where it helps the values that would have been accepted.

Code

What happened

What to do

not_found

Lever answered, and holds no such site or opening.

Check the site name with resolve_company.

invalid_input

The arguments were refused before any request went out.

Read the message, which names the argument and what it takes.

rate_limited

Lever asked this client to slow down.

Wait, then call again with the same arguments. The opening is still on the board.

parse_failure

Lever answered in a shape this client cannot read.

Report it at the issue tracker.

network_error

The request did not complete.

Try again shortly.

timeout

The request passed its deadline.

Ask for fewer companies, or a smaller limit.

As a library

The layer reading Lever is published on its own, with its pacing, its cache and its errors, and with no protocol attached.

import { Client } from "mcp-lever/client";

const client = new Client({ minIntervalMs: 2000 });
const resolved = await client.resolveCompany("Included Health");
const jobs = await client.listPostings(resolved.found[0], { limit: 10 });
console.log(jobs.length);

ClientOptions takes minIntervalMs, timeoutMs, cacheTtlMs and fetchImpl. An interval below the published floor is ignored, so the floor holds here as well.

Pacing and attribution

Both API hosts publish Crawl-delay: 1, so requests go out one at a time with at least a second between them, and that floor holds however the client is configured. The User-Agent carries the project and an address where a person can be reached, and imitates no browser.

Reads go to api.lever.co and api.eu.lever.co, which are the hosts Lever documents for its posting data. The jobs.lever.co careers pages are left alone.

Every opening carries the address of its Lever page and its apply URL. Credit the company and link that page when you show an opening.

This MCP server is an unofficial project, with no affiliation to Lever or to the companies whose boards it reads.

Privacy

This server collects nothing about you and sends nothing to its author. It runs on your machine, contacts api.lever.co and api.eu.lever.co and nothing else, holds its answers in memory while it runs, and writes nothing to disk. PRIVACY.md states what a request carries and which settings change any of it.

Development

npm install
npm run build:fixtures
npm test
npm run check

Tests run against generated fixtures and make no network request. The live suite, npm run test:live, makes one request per route and runs nightly against the service itself.

Contributing

Bugs, questions and ideas belong in the issue tracker. Pull requests are welcome; opening an issue first helps agree on the shape of the change. See CONTRIBUTING.md.

License

MIT, see LICENSE. The openings belong to the companies that published them.


mcp-lever (français)

English version

Lever est un logiciel de recrutement qu'utilisent des milliers d'entreprises pour mener leurs embauches, et chaque cliente reçoit avec lui un site d'offres public. Chaque site porte les postes ouverts de cette entreprise avec leur intitulé, leur lieu, l'équipe et le département auxquels ils appartiennent, le type de contrat demandé, l'annonce complète, et la fourchette de salaire quand l'entreprise a choisi d'en publier une. Lever héberge un site par entreprise, sur son instance mondiale ou sur son instance européenne, et ne publie aucun index les traversant.

Ce serveur relie un client de conversation à ces sites. Vous nommez les entreprises qui vous intéressent, et il traduit chaque nom en l'identifiant qui adresse son site, cherche dans leurs offres, les filtre par lieu, équipe, mode de travail, pays, salaire ou fraîcheur de publication, lit une offre en entier, et liste les formulations selon lesquelles chaque entreprise classe les siennes. Aucune clé d'API, aucun compte.

Installation

Installation en un clic

Install in Cursor Install in VS Code

Claude Code

claude mcp add lever -- npx -y mcp-lever

Claude Desktop, Cursor, et tout client au format de configuration standard

{
  "mcpServers": {
    "lever": {
      "command": "npx",
      "args": ["-y", "mcp-lever"]
    }
  }
}

Node 24 ou plus récent est nécessaire, et aucune variable d'environnement n'est à renseigner.

Avec Docker

{
  "mcpServers": {
    "lever": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "ghcr.io/smeet666/mcp-lever:2.1.0"]
    }
  }
}

-i garde l'entrée standard ouverte, qui est le canal du protocole, et -t est omis parce qu'un TTY réécrit le flux. Le conteneur a besoin d'un accès HTTPS sortant vers api.lever.co et api.eu.lever.co, et de rien d'autre : aucun volume, aucun port, aucun identifiant.

Bundle, sans npm

Téléchargez mcp-lever-2.1.0.mcpb depuis la dernière publication et ouvrez-le. Un client qui gère les bundles MCP l'installe seul, sans npm et sans fichier de configuration à modifier. Le bundle emporte ses dépendances, donc rien n'est téléchargé à l'installation.

Ce qu'on peut demander

  • « Lesquelles d'Included Health, Netlify et Ramp recrutent sur Lever ? »

  • « Trouve-moi des postes d'ingénierie en télétravail chez ces trois-là. »

  • « Lis-moi cette offre en entier. »

  • « Sous quels lieux Included Health classe-t-elle ses offres ? »

  • « Quelque chose publié ces quinze derniers jours chez Netlify ? »

Chaque question part d'une entreprise, puisque Lever n'offre aucune recherche traversant les sites. search_jobs résout lui-même les noms qu'on lui donne, donc rien n'est à préparer :

resolve_company(["Included Health"])  ->  includedhealth, instance mondiale, publie
search_jobs(["Included Health"], keyword: "therapist")
get_job("includedhealth", "6f97a19f-…")

Les outils

Outil

Ce qu'il fait

resolve_company

Traduit des noms d'entreprises en identifiants de sites Lever.

search_jobs

Cherche dans les offres des entreprises nommées.

get_job

Lit une offre en entier, annonce comprise.

list_filter_values

Liste les formulations sous lesquelles une entreprise classe.

Un identifiant de site Lever distingue la casse, donc Flex répond là où flex ne rend rien. Quatre orthographes sont essayées par nom sur chacune des deux instances, et la réponse liste ce qui a été envoyé : ne rien trouver ne prouve jamais qu'une entreprise est absente de Lever.

resolve_company

Traduit des noms d'entreprises en identifiants de sites Lever, en signalant chaque instance qui a répondu. Il prend une liste.

Argument

Type

Requis

Ce qu'il fait

names

tableau de 1 à 25 chaînes

oui

Des noms d'entreprises, ou des identifiants déjà connus.

En retour : une entrée par nom, portant input ; found, une liste de { slug, instance, publishes } où publishes est faux pour un site qui existe et ne liste rien aujourd'hui ; tried, les orthographes envoyées dans l'ordre ; et cached, vrai quand la session avait déjà résolu ce nom. Un nom qui répond sur les deux instances revient avec les deux, et aucune n'est élue : passez celle que vous visez aux autres outils.

search_jobs

Cherche dans les offres des entreprises nommées. Lever applique les filtres qu'il gère sur sa propre formulation exacte, et ce serveur applique les autres aux offres qu'il a lues.

Argument

Type

Requis

Ce qu'il fait

companies

tableau de 1 à 25 chaînes

oui

Noms d'entreprises ou identifiants. Chacun est résolu ici.

keyword

chaîne

non

Mots à chercher dans l'intitulé et dans l'annonce.

location

tableau de 1 à 20 chaînes

non

Des lieux, exactement comme Lever les écrit.

team

tableau de 1 à 20 chaînes

non

Des équipes, exactement comme Lever les écrit.

department

tableau de 1 à 20 chaînes

non

Des départements, exactement comme Lever les écrit.

commitment

tableau de 1 à 20 chaînes

non

Des types de contrat, exactement comme Lever les écrit.

workplace_type

tableau de 1 à 4 chaînes

non

remote, hybrid, onsite ou unspecified.

country

tableau de 1 à 20 codes à deux lettres

non

Des pays en code ISO, comme FR ou US.

salary_min

nombre, 0 ou plus

non

La plus basse borne haute de fourchette à conserver.

salary_interval

chaîne

non

La période dans laquelle salary_min est écrit, par exemple per-year-salary.

currency

code à trois lettres

non

La devise dans laquelle salary_min est écrit, comme EUR.

posted_within_days

entier, 1 à 3650

non

L'ancienneté maximale d'une offre.

limit

entier, 1 à 100, défaut 25

non

Offres à lire par entreprise.

skip

entier, 0 à 100000, défaut 0

non

Offres à enjamber par entreprise.

Lever applique lui-même location, team, department et commitment ; ce serveur applique keyword, workplace_type, country, salary_min, salary_interval, currency et posted_within_days à ce qu'il a lu. list_filter_values publie les formulations que prennent les quatre premiers, et une formulation que Lever ignore revient en liste vide.

En retour : jobs, chacune portant id et company_slug, que get_job reprend, plus title, location, all_locations, country, workplace_type, team, posted_at, url et apply_url. commitment et department sont absents quand l'entreprise ne les renseigne pas. salary vaut null pour une offre publiée sans fourchette, ce qui ne vaut jamais zéro, et porte l'interval dans lequel Lever l'a écrite, jamais converti ni annualisé. per_company donne une issue par entreprise, avec un status valant read, unresolved, empty ou failed, qui sont quatre réponses différentes, et les comptes read et returned de part et d'autre des filtres. total_available vaut toujours null : Lever ne publie aucun compte de résultats. Les lignes ne portent pas l'annonce, un site d'entreprise pouvant peser plusieurs mégaoctets.

limit s'applique par entreprise, et une entreprise dont les offres le remplissent en publie peut-être davantage : les notes le signalent, et disent qu'un compte pris dans cette fenêtre mesure la fenêtre. posted_within_days parcourt jusqu'à cinq pages par entreprise, et Lever pagine par intitulé, donc une offre publiée hier peut se trouver n'importe où dans un site.

get_job

Lit une offre en entier : l'annonce, ses sections nommées, et le salaire tel que publié.

Argument

Type

Requis

Ce qu'il fait

company_slug

chaîne

oui

L'identifiant du site, rendu par resolve_company.

job_id

chaîne

oui

L'identifiant d'une offre, rendu par une recherche.

instance

global ou eu

non

L'instance d'où vient la ligne. La mondiale par défaut.

En retour : job, qui porte les champs d'une ligne de recherche, plus description, sections en { heading, items }, salary_note pour ce que l'entreprise a écrit à côté de la fourchette, et source avec l'adresse d'où l'offre a été lue.

list_filter_values

Liste les formulations d'équipe, de lieu et de contrat qu'une entreprise emploie. À lire avant de filtrer : Lever fait correspondre sa propre formulation, et le vocabulaire appartient à chaque entreprise, l'une écrivant Full-time là où une autre écrit EE Full-Time.

Argument

Type

Requis

Ce qu'il fait

company_slug

chaîne

oui

L'identifiant du site, rendu par resolve_company.

instance

global ou eu

non

L'instance où vit ce site. La mondiale par défaut.

fields

tableau de 1 à 3 parmi team, location, commitment

non

Les vocabulaires à lire. Chacun coûte une requête, et les trois sont lus par défaut.

En retour : company_slug, instance, et fields qui porte une liste de { value, count } pour chaque vocabulaire demandé. Un count vaut null là où Lever n'a publié aucun chiffre à côté de la catégorie.

Configuration

Il n'y a rien à configurer. Le serveur ne lit aucune variable d'environnement, et le bloc mcpServers ci-dessus est complet tel quel.

Le rythme, le délai et le cache sont des réglages de la couche cliente, que Comme bibliothèque montre comment passer. L'écart entre deux requêtes peut y être élargi et jamais resserré.

Erreurs

Chaque échec porte un des six codes, un message, et quand cela aide les valeurs qui auraient été acceptées.

Code

Ce qui s'est passé

Que faire

not_found

Lever a répondu, et n'a ni ce site ni cette offre.

Vérifiez l'identifiant avec resolve_company.

invalid_input

Les arguments ont été refusés avant toute requête.

Lisez le message, qui nomme l'argument et ce qu'il prend.

rate_limited

Lever demande à ce client de ralentir.

Attendez, puis rappelez avec les mêmes arguments. L'offre est toujours en ligne.

parse_failure

Lever a répondu dans une forme que ce client ne lit pas.

Signalez-le sur le suivi d'incidents.

network_error

La requête n'a pas abouti.

Réessayez sous peu.

timeout

La requête a dépassé son délai.

Demandez moins d'entreprises, ou un limit plus petit.

Comme bibliothèque

La couche qui lit Lever est publiée seule, avec son rythme, son cache et ses erreurs, sans protocole attaché.

import { Client } from "mcp-lever/client";

const client = new Client({ minIntervalMs: 2000 });
const resolved = await client.resolveCompany("Included Health");
const jobs = await client.listPostings(resolved.found[0], { limit: 10 });
console.log(jobs.length);

ClientOptions prend minIntervalMs, timeoutMs, cacheTtlMs et fetchImpl. Un écart sous le plancher publié est ignoré, donc le plancher tient également ici.

Rythme et attribution

Les deux hôtes d'API publient Crawl-delay: 1, donc les requêtes partent une à une avec au moins une seconde entre elles, et ce plancher tient quelle que soit la configuration du client. Le User-Agent porte le projet et une adresse où joindre une personne, et n'imite aucun navigateur.

Les lectures vont vers api.lever.co et api.eu.lever.co, les hôtes que Lever documente pour ses données d'offres. Les pages carrières jobs.lever.co sont laissées tranquilles.

Chaque offre porte l'adresse de sa page Lever et son adresse de candidature. Créditez l'entreprise et renvoyez vers cette page quand vous montrez une offre.

Ce MCP est un projet non officiel, sans affiliation à Lever ni aux entreprises dont il lit les sites.

Confidentialité

Ce serveur ne collecte rien sur vous et n'envoie rien à son auteur. Il tourne sur votre machine, ne joint que api.lever.co et api.eu.lever.co, garde ses réponses en mémoire le temps qu'il tourne, et n'écrit rien sur le disque. PRIVACY.md dit ce qu'une requête emporte et quels réglages changent cela.

Développement

npm install
npm run build:fixtures
npm test
npm run check

Les tests s'exécutent sur des fixtures engendrées et n'émettent aucune requête. La suite en direct, npm run test:live, émet une requête par route et tourne chaque nuit contre le service lui-même.

Contribuer

Les anomalies, les questions et les idées ont leur place dans le suivi d'incidents. Les propositions de modification sont bienvenues ; ouvrir un ticket d'abord aide à s'accorder sur la forme du changement. Voir CONTRIBUTING.md.

Licence

MIT, voir LICENSE. Les offres appartiennent aux entreprises qui les ont publiées.

Available Tools

4 tools
get_jobRead one openingA
Read-onlyIdempotent

Read one Lever opening in full: the advert, its named sections, and the salary as published. company_slug and instance are the ones the row carried, and a site living on both instances holds different openings on each.

NameRequiredDescriptionDefault
job_idYesThe identifier of one opening, as a search returns it.
instanceNoThe Lever instance the row came from. Left out, the global one is read.
company_slugYesThe Lever site name, as resolve_company returns it.

Output Schema

NameRequiredDescription
jobYes
notesYes

TDQS

A4.2/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare read-only, idempotent, open-world, and non-destructive behavior. The description adds meaningful context beyond that: it describes what the read returns and warns that a site present on both instances has different openings on each, which prevents cross-instance mistakes.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two focused sentences with no filler. The core purpose is front-loaded, and the instance distinction is stated compactly afterward.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With annotations covering safety hints and an output schema covering return shape, the description supplies enough to call this correctly, including the key instance caveat. Minor omissions such as not-found behavior or explicit routing from search_jobs are not critical for a simple read tool.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the schema already defines all three parameters. The description still adds semantic nuance by explaining that company_slug and instance should match the originating row and that the same site can have distinct openings per instance, which is not captured in the schema alone.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description names a specific verb and resource: reading one Lever opening in full. It also enumerates the returned content (advert, named sections, published salary), which distinguishes this from search_jobs and makes the tool's purpose unambiguous.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It implies use after obtaining a row from search or resolution by saying company_slug and instance are those the row carried, but it never explicitly names search_jobs or states when not to use this tool. The instance caveat is useful, but alternative-routing guidance remains implicit.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

list_filter_valuesList a company's filter wordingsA
Read-onlyIdempotent

List the team, location and commitment wordings one company uses on Lever, so a filter can carry the words that company publishes. The vocabulary belongs to each company: one writes Full-time, another Full Time, another EE Full-Time. Lever publishes no vocabulary for department, so a department filter is checked against the openings a company has open right now.

NameRequiredDescriptionDefault
fieldsNoWhich vocabularies to read. Each costs one request. All three by default.
instanceNoThe Lever instance this site lives on. Left out, the global one is read.
company_slugYesThe Lever site name, as resolve_company returns it.

Output Schema

NameRequiredDescription
notesYes
fieldsYes
instanceYes
company_slugYes

TDQS

A4.3/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already cover readOnlyHint, openWorldHint, idempotentHint, and destructiveHint. The description adds meaningful behavioral context: vocabulary is company-specific (e.g., 'Full-time' vs 'Full Time') and department filters are checked against current openings because no vocabulary is published. This goes beyond the annotations and helps the agent understand the data's nature.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences, front-loaded with the action, and efficiently explains the tool's purpose, the per-company variation, and the department exception. Every sentence adds valuable context without redundancy.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a simple read-only tool with an output schema present, the description sufficiently covers what the tool does, why it is used, and an important edge case (department). The combination of annotations, schema descriptions, and the description gives an agent everything needed to select and invoke the tool correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The input schema describes all parameters with 100% coverage, including defaults (fields defaults to all three, instance defaults to global, company_slug is required). The description underscores the purpose of fields (team, location, commitment) but adds no new parameter-level syntax or details. Baseline 3 is appropriate when schema carries the semantic load.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's function: 'List the team, location and commitment wordings one company uses on Lever.' It names the specific verb ('List'), the resource (filter wordings for a company), and the scope (team, location, commitment), distinguishing it from sibling tools like search_jobs and get_job.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explains when this tool is valuable: when building filters that need company-specific wording ('so a filter can carry the words that company publishes'). It also provides context by noting that department vocabulary is not published and is handled differently, implicitly guiding the agent away from expecting it. It does not explicitly name alternatives but the use case is clear.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

resolve_companyResolve company names to Lever boardsA
Read-onlyIdempotent

Turn company names into the Lever site names that address their job boards, and report every instance that answered. search_jobs resolves names on its own, so reach for this to check which of several companies are on Lever at all: it probes names without reading a single board, where a search would read one per company.

NameRequiredDescriptionDefault
namesYesA list of a company name, or a Lever site name you already know, one to 25 of them.

Output Schema

NameRequiredDescription
notesYes
resolvedYes

TDQS

A4.5/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnlyHint, openWorldHint, idempotentHint, and destructiveHint. The description adds valuable context: it reports every instance that answered and probes names without reading a single board, clarifying the operational behavior and lightweight nature beyond the annotations.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences, with the first sentence stating the core function and the second providing usage guidance. Every sentence contributes meaning, with no fluff or repetition of structured data.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the simple parameter schema, comprehensive annotations, and presence of an output schema, the description covers all necessary aspects: purpose, usage differentiation, and behavioral traits. It does not need to explain return values because an output schema exists.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The single parameter 'names' is fully described in the schema (100% coverage), so the description does not need to add parameter details. It aligns with the schema by referring to 'company names' and 'Lever site names', but adds no additional syntax or format information, matching the baseline for full schema coverage.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description uses a specific verb ('turn') and resource ('company names' to 'Lever site names'), clearly stating the mapping action and output. It also distinguishes itself from search_jobs by noting that this tool checks which companies are on Lever without reading boards.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly states when to use this tool versus search_jobs: search_jobs resolves names on its own, while this tool is for checking which of several companies are on Lever. It also explains the efficiency advantage (probes without reading boards).

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

search_jobsSearch openings at named companiesA
Read-onlyIdempotent

Search the openings published by named companies on Lever. companies takes company names or Lever site names, and each name is turned into a site name here. A name found on the first spelling costs 3 requests, and one that resists costs up to 9, at one second each. location, team, department and commitment are sent to Lever and need its exact wording. keyword, workplace_type, country, salary and recency are applied here, because Lever accepts none of them as filters. limit applies per company, so a company whose openings fill it may publish more. posted_within_days walks up to 5 pages per company, because Lever pages by title and a recent opening sits anywhere.

NameRequiredDescriptionDefault
skipNoHow many openings to step over per company.
teamNoA list of a team exactly as Lever writes it, one to 20 of them.
limitNoHow many openings to read per company.
countryNoA list of a two-letter country code, as in FR or US, one to 20 of them.
keywordNoWords to look for in the title and the advert.
currencyNoA three-letter currency code, as in EUR or USD.
locationNoA list of a location exactly as Lever writes it, one to 20 of them.
companiesYesA list of a company name or a Lever site name, one to 25 of them.
commitmentNoA list of a commitment exactly as Lever writes it, one to 20 of them.
departmentNoA list of a department exactly as Lever writes it, one to 20 of them.
salary_minNoThe lowest upper bound of a salary range to keep.
workplace_typeNoA list of remote, hybrid, onsite or unspecified, one to 4 of them.
salary_intervalNoThe period salary_min is written in, as Lever writes it, such as per-year-salary or per-hour-wage.
posted_within_daysNoHow recent an opening must be.

Output Schema

NameRequiredDescription
jobsYes
notesYes
per_companyYes
total_availableYesLever publishes no result count, so this is always null.

TDQS

A4.1/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Goes well beyond the annotations by disclosing request cost (3 requests for a clean company name, up to 9 at one second each for a stubborn one), per-company limit semantics (a company that fills the limit may publish more), and the up-to-5-page walk for posted_within_days because Lever pages by title. These are exactly the operational traits an agent needs before committing to a call.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the purpose sentence, then each subsequent sentence attaches to a specific concern (companies mapping, cost, server-vs-local filters, limit, recency paging). No filler and no repetition of the schema text.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 14-parameter tool with an output schema, it covers the hard parts an agent would otherwise get wrong: cost, filter routing, per-company limits and the recency page walk. Minor gaps remain around how skip interacts with per-company paging and what happens when a company name cannot be resolved.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the baseline is 3; the description earns above that by adding semantics the schema does not carry: company names are converted to Lever site names internally, and location/team/department/commitment are server-side while keyword, workplace_type, country, salary and recency are filtered locally. It leaves skip and salary_min/salary_interval to the schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource: it searches openings published by named companies on Lever, with a clear companies-driven scope. It is distinguishable from get_job (single opening) and resolve_company (name resolution), though it does not name those siblings to draw the boundary explicitly.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explains which filters Lever accepts versus which are applied locally, and warns about per-company cost, which is useful invocation guidance. However, it never says when to use this versus resolve_company or list_filter_values, and the 'need its exact wording' note would be stronger if it pointed at list_filter_values as the way to obtain that wording.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 2 tool updatesv2.1.0
    • Changedget_job4 fields changed
      • removedOutput schema / properties / job / properties / country / anyOf
        Removed value: -[
        -  {
        -    "type": "string"
        -  },
        -  {
        -    "type": "null"
        -  }
        -]
      • addedOutput schema / properties / job / properties / country / type
        Added value: +[
        +  "string",
        +  "null"
        +]
      • removedOutput schema / properties / job / properties / salary_note / anyOf
        Removed value: -[
        -  {
        -    "type": "string"
        -  },
        -  {
        -    "type": "null"
        -  }
        -]
      • addedOutput schema / properties / job / properties / salary_note / type
        Added value: +[
        +  "string",
        +  "null"
        +]
    • Changedsearch_jobs4 fields changed
      • removedOutput schema / properties / jobs / items / properties / country / anyOf
        Removed value: -[
        -  {
        -    "type": "string"
        -  },
        -  {
        -    "type": "null"
        -  }
        -]
      • addedOutput schema / properties / jobs / items / properties / country / type
        Added value: +[
        +  "string",
        +  "null"
        +]
      • removedOutput schema / properties / per_company / items / properties / slug / anyOf
        Removed value: -[
        -  {
        -    "type": "string"
        -  },
        -  {
        -    "type": "null"
        -  }
        -]
      • addedOutput schema / properties / per_company / items / properties / slug / type
        Added value: +[
        +  "string",
        +  "null"
        +]
  2. 4 tool updatesv1.0.0
    • First observedget_job
    • First observedlist_filter_values
    • First observedresolve_company
    • First observedsearch_jobs

TDQS

A4.3/5.0

Scored across 4 tools

Disambiguation4/5

search_jobs and resolve_company share the company-name-resolution step, but the descriptions clearly separate probing which companies are on Lever from actually reading job boards. get_job and list_filter_values have distinct purposes. There is still a small moment of hesitation over whether to call resolve_company first, but otherwise the tools are easy to tell apart.

Naming Consistency5/5

All four tools use a consistent snake_case verb_noun pattern: search_jobs, resolve_company, get_job, list_filter_values. The convention is predictable and readable throughout.

Tool Count5/5

Four tools is a well-scoped set for a read-only Lever job-search server. Each tool earns its place: search, company resolution, job detail, and filter vocabulary lookup.

Completeness4/5

The read-only lifecycle is covered: resolving company names, searching openings, reading a full job, and listing filter vocabularies. Minor gaps remain, such as no direct department-vocabulary listing and no way to browse all Lever companies without naming them, but agents can work around these.

Maintenance

ActivityActive
ResponsivenessUnresponsive

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    A
    maintenance
    MCP server that exposes job search data from multiple boards, enabling clients to query and manage job listings via natural language.
    7
    MIT
  • A
    license
    Not graded
    quality
    D
    maintenance
    Unified job search MCP server that aggregates live listings from multiple job boards with deduplication, enabling AI agents to find and filter jobs by keyword and location.
    MIT
  • A
    license
    A
    quality
    D
    maintenance
    MCP server that scours job openings from public, ToS-clean sources (Greenhouse, Lever, Ashby, HN, RemoteOK, Adzuna, USAJobs) and provides tools for job search, company listings, and salary context.
    4
    MIT
  • A
    license
    Not graded
    quality
    D
    maintenance
    MCP server that aggregates and deduplicates job listings from multiple public sources, ranks them against a user's resume, and exposes tools for searching, viewing details, explaining fit, and tracking applications.
    MIT