This server lets you browse tvsubtitles.net's catalogue of TV series subtitles through MCP tools: search for shows, inspect season and episode coverage, read subtitle records, and list available languages.
search_titles: Find a television series by name, optionally filter by year, limit results, and get show IDs for other tools.
list_subtitles: View a season's coverage (which languages exist per episode) or list all subtitle records for a specific episode, with optional language filtering.
get_subtitle: Fetch detailed metadata for one subtitle record, including release info, uploader, download page URL, file size, rating, and publish date.
list_languages: List the site's supported languages and their codes, or see which languages a specific show/season actually has subtitles in.
All tools are read-only, require no API key or account, and return source URLs so you can link to the download pages.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@mcp-tvsubtitles Do you have French subtitles for Smallville?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
TVsubtitles.net catalogues subtitles for television series. Its readers upload a file for one episode, timed to one video release, and the site records the language, the release it was cut for, who uploaded it and when, how large the file is and how often it has been taken. It holds around three hundred thousand of them, across some eighty-six thousand episodes in twenty-four languages.
This server connects a chat client to that catalogue. You can search for a series, read a season and see which languages hold something for each episode, read the records of one episode, and open one record with the release it was cut for and who uploaded it. Every record carries the address of its page on the site. No API key and no account are needed.
One-click install
Claude Code
claude mcp add tvsubtitles -- npx -y mcp-tvsubtitlesAny client reading mcpServers
{
"mcpServers": {
"tvsubtitles": {
"command": "npx",
"args": ["-y", "mcp-tvsubtitles"]
}
}
}{
"mcpServers": {
"tvsubtitles": {
"command": "docker",
"args": ["run", "--rm", "-i", "ghcr.io/smeet666/mcp-tvsubtitles:1.0.0"]
}
}
}The container reaches https://www.tvsubtitles.net and nothing else.
Download mcp-tvsubtitles.mcpb from the
latest release and
open it with a host that installs MCP bundles. It carries its dependencies, so
Node 24 or later is all it needs.
Related MCP server: OpenSubtitles MCP Server
"Does tvsubtitles have French subtitles for Smallville?"
"Which languages cover season 3 of Harbour Lights?"
"List the English subtitles for episode 7 of that season."
"Which release is that subtitle timed to, and who uploaded it?"
"Find me the page to download the Polish subtitle for the finale."
Tool | What it does |
| Finds a television series by name, and returns the id the others take. |
| Reads a season's coverage, or one episode's subtitle records. |
| Reads one record, with its release, its uploader and its download page. |
| Lists the languages the catalogue holds, or the ones one show holds. |
search_titlesSearches the catalogue by name. The site catalogues television only, so a search for a film is refused rather than answered.
Argument | Type | Required | What it does |
| string, 1–120 characters | yes | The name of a series, or part of one. |
|
| no |
|
| integer, 1900–2100 | no | Keeps rows whose published years cover this one. |
| integer, 1–100, default 20 | no | Rows to render. |
| boolean, default false | no | Reads each row's subtitle, episode and season counts from the catalogue index, at the cost of one further request. |
In return: the matching series, each with the id the other tools take, the
years the site publishes, and the languages it draws a flag for. imdb_id and
tmdb_id are null, because the site's search publishes none. subtitle_count,
episode_count and season_count are null unless with_counts asks for them,
since the site publishes those three on its catalogue index rather than on the
page a search answers with; each counts every season of the show together, which
counts_scope names, each is read on its own, and a show the index carries no
row for keeps three nulls. total_available counts the rows this one search came
back with, which total_counts names. A year that leaves nothing is set aside
and named in filters_dropped.
list_subtitlesReads what a series holds, in one of two shapes that kind names.
Argument | Type | Required | What it does |
| string, 1–12 characters | yes | A show id from |
| integer, 1–200 | no | Left out, the newest season the site holds is read. |
| integer, 1–500 | no | Named, the answer is that episode's records. |
| string, 1–40 characters | no | A language from |
| integer, 1–200, default 40 | no | Rows to render. |
In return: with a season alone, kind reads coverage and each row is one
episode, carrying episode_id, the number of subtitles the site counts and the
languages holding something. With an episode named, kind reads subtitles and
each row is a record carrying the id that get_subtitle takes. season is the
season the site served and season_requested the one that was asked for, which
differ when the newest was read. seasons_available lists the seasons the show
holds. A language holding nothing is set aside, the answer comes back
unnarrowed, and filters_dropped names it.
get_subtitleReads one record from an id list_subtitles returned.
Argument | Type | Required | What it does |
| string, 1–12 characters | yes | A subtitle id from |
In return: the record, with page_url for the page a reader opens to
download the file. read_from reads record here and listing on a row
list_subtitles produced, which is what tells an unread field from one the site
does not publish: a listing carries no file_name, no size_text and no
comment, because the site prints those on a record's own page alone. releases holds the video releases the site published and
release_match says whether it published any: stated where it did, none
where it did not, so a record marked none says nothing about which video it is
timed to. uploader is null on roughly two records out of three. published_at
is the site's stamp read into ISO 8601 and carries no timezone, and
published_text keeps the site's own wording. rating holds two counters the
site publishes, where a zero is a figure it printed.
list_languagesLists the languages the site catalogues, or the ones one show holds.
Argument | Type | Required | What it does |
| string, 1–12 characters | no | A show id, to read what that show holds. |
| integer, 1–200 | no | Which season to measure a show over. Ignored without |
In return: each language with the name the site prints, the two-letter
site_code it addresses the language by, the BCP 47 code where that mapping is
certain, and differs_from_iso. scope says what was measured: catalogue for
the whole site, or season when a show id was passed, and count is then the
episodes of that season holding the language.
The site draws twenty-four flags and addresses each language by two letters of
its own choosing. Six of them differ from ISO 639-1, and one collides: the site
writes br for Brazilian Portuguese, which ISO assigns to Breton.
Site code | Language | BCP 47 |
| Brazilian Portuguese |
|
| Greek |
|
| Czech |
|
| Japanese |
|
| Chinese |
|
| Ukrainian |
|
language keeps the site's own name and language_code carries the tag, so
nothing has to be derived from the two letters. list_subtitles accepts a
language written any of the three ways.
Nothing has to be set. Every variable below is optional.
Variable | Default | Range | What it does |
| unset | Prepended to this server's own agent, which stays attached. | |
| 2000 | 1500–60000 | Milliseconds between two requests. 1500 is a floor. |
| 20000 | 1000–120000 | Deadline for one attempt. |
| 60000 | 5000–600000 | Deadline for one read, its retries included. |
| 3 | 0–8 | Attempts after the first. |
| 900000 | 0–86400000 | How long a page is held in memory. 0 turns the store off. |
| 200 | 1–5000 | Pages held before the least recently used is dropped. |
| 8000000 | 100000–64000000 | The largest answer read for one page. |
|
|
| What reaches stderr. |
A value outside its range is refused on stderr and the default stands.
Code | What it means | What to do |
| The site answered, and holds no such thing. | Check the id came from a listing rather than by hand. |
| The arguments could not produce a request. | Read the message, which names the argument. |
| The site asked this client to slow down. | Wait and ask again. The thing asked for still exists. |
| An answer arrived in a shape this cannot read. | Report it with the arguments used. |
| The request did not complete. | Try again. |
| No answer arrived inside the deadline. | Try again, or raise |
The layer that reads the site is published on its own, with the pacing, the store and the error codes, and no protocol attached.
import { TvSubtitlesClient } from "mcp-tvsubtitles/client";
const client = new TvSubtitlesClient();
const found = await client.searchShows("Smallville");
const season = await client.getSeason(found.data.rows[0].id, 0);
console.log(season.data.showName, season.data.season, season.data.episodes.length);Every read returns { data, cached }, with skipped when rows were left out.
The constructor takes { config, logger, fetchImpl }, and the interval floor
holds whatever is passed.
One request at a time, two seconds apart, widening when the site pushes back and narrowing again on a run of clean answers. The floor of 1.5 seconds cannot be lowered. The agent string carries the project name, its version and the address of this repository, so the site can reach a person.
Subtitles are the work of the people who wrote and timed them. This server reads
the catalogue and downloads no subtitle file: every record carries page_url,
which is the page a reader opens to download it. Credit tvsubtitles.net and link
that page when you show a result.
This MCP server is not affiliated with tvsubtitles.net.
No account, no key, no telemetry. The only host reached is
https://www.tvsubtitles.net. Pages are held in memory for fifteen minutes and
nothing is written to disk. Diagnostics go to stderr. See PRIVACY.md.
npm install
npm run build:fixtures
npm test
npm run coverage
npm run checknpm run test:live makes one request per route against the site, and runs
nightly.
Issues and pull requests are welcome. See CONTRIBUTING.md and SECURITY.md.
MIT, see LICENSE.
TVsubtitles.net catalogue les sous-titres de séries télévisées. Ses lecteurs y déposent un fichier pour un épisode, calé sur une version vidéo, et le site enregistre la langue, la version pour laquelle il a été taillé, qui l'a déposé et quand, la taille du fichier et le nombre de fois qu'il a été pris. Il en tient environ trois cent mille, sur quelque quatre-vingt-six mille épisodes, dans vingt-quatre langues.
Ce serveur relie un client de conversation à ce catalogue. On peut chercher une série, lire une saison en voyant quelles langues tiennent quelque chose pour chaque épisode, lire les fiches d'un épisode, et ouvrir une fiche avec la version pour laquelle elle a été taillée et qui l'a déposée. Chaque fiche porte l'adresse de sa page sur le site. Aucune clé d'API ni aucun compte ne sont nécessaires.
Installation en un clic
Claude Code
claude mcp add tvsubtitles -- npx -y mcp-tvsubtitlesTout client lisant mcpServers
{
"mcpServers": {
"tvsubtitles": {
"command": "npx",
"args": ["-y", "mcp-tvsubtitles"]
}
}
}{
"mcpServers": {
"tvsubtitles": {
"command": "docker",
"args": ["run", "--rm", "-i", "ghcr.io/smeet666/mcp-tvsubtitles:1.0.0"]
}
}
}Le conteneur joint https://www.tvsubtitles.net et rien d'autre.
Télécharger mcp-tvsubtitles.mcpb depuis la
dernière publication
et l'ouvrir avec un hôte qui installe les bundles MCP. Il emporte ses
dépendances, donc Node 24 ou plus récent suffit.
« Est-ce que tvsubtitles a des sous-titres français pour Smallville ? »
« Quelles langues couvrent la saison 3 de Harbour Lights ? »
« Liste les sous-titres anglais de l'épisode 7 de cette saison. »
« Sur quelle version ce sous-titre est-il calé, et qui l'a déposé ? »
« Trouve-moi la page pour télécharger le sous-titre polonais du dernier épisode. »
Outil | Ce qu'il fait |
| Trouve une série par son nom et rend l'id que les autres prennent. |
| Lit la couverture d'une saison, ou les fiches d'un épisode. |
| Lit une fiche, avec sa version, son déposant et sa page de téléchargement. |
| Liste les langues du catalogue, ou celles que tient une série. |
search_titlesCherche dans le catalogue par le nom. Le site ne catalogue que la télévision, donc une recherche de film est refusée au lieu d'être répondue.
Argument | Type | Requis | Ce qu'il fait |
| chaîne, 1 à 120 caractères | oui | Le nom d'une série, ou une partie. |
|
| non |
|
| entier, 1900 à 2100 | non | Garde les lignes dont les années publiées couvrent celle-ci. |
| entier, 1 à 100, défaut 20 | non | Lignes à rendre. |
| booléen, défaut faux | non | Lit les comptes de sous-titres, d'épisodes et de saisons de chaque ligne dans l'index du catalogue, au prix d'une requête de plus. |
En retour : les séries trouvées, chacune avec l'id que prennent les autres
outils, les années publiées par le site et les langues pour lesquelles il dessine
un drapeau. imdb_id et tmdb_id valent null, puisque la recherche du site n'en
publie aucun. subtitle_count, episode_count et season_count valent null
tant que with_counts ne les demande pas, car le site publie ces trois chiffres
sur l'index de son catalogue et non sur la page qui répond à une recherche ;
chacun compte toutes les saisons de la série ensemble, ce que counts_scope
nomme, chacun se lit à part, et une série dont l'index ne porte aucune ligne
garde trois null. total_available compte les lignes que cette recherche a
ramenées, ce que total_counts nomme. Une année qui ne laisse rien est mise de
côté et nommée dans filters_dropped.
list_subtitlesLit ce qu'une série tient, sous l'une de deux formes que kind nomme.
Argument | Type | Requis | Ce qu'il fait |
| chaîne, 1 à 12 caractères | oui | Un id de série venu de |
| entier, 1 à 200 | non | Omis, la saison la plus récente est lue. |
| entier, 1 à 500 | non | Nommé, la réponse porte les fiches de cet épisode. |
| chaîne, 1 à 40 caractères | non | Une langue de |
| entier, 1 à 200, défaut 40 | non | Lignes à rendre. |
En retour : avec une saison seule, kind vaut coverage et chaque ligne est
un épisode, portant episode_id, le nombre de sous-titres que le site compte et
les langues qui tiennent quelque chose. Avec un épisode nommé, kind vaut
subtitles et chaque ligne est une fiche portant l'id que prend
get_subtitle. season est la saison servie par le site et season_requested
celle demandée, les deux différant quand la plus récente a été lue.
seasons_available liste les saisons que la série tient. Une langue qui ne tient
rien est mise de côté, la réponse revient sans la restriction, et
filters_dropped la nomme.
get_subtitleLit une fiche depuis un id rendu par list_subtitles.
Argument | Type | Requis | Ce qu'il fait |
| chaîne, 1 à 12 caractères | oui | Un id de sous-titre venu de |
En retour : la fiche, avec page_url pour la page qu'un lecteur ouvre afin
de télécharger le fichier. read_from vaut ici record, et listing sur une
ligne venue de list_subtitles : c'est ce qui distingue un champ non lu d'un
champ que le site ne publie pas, car une liste ne porte ni file_name, ni
size_text, ni comment, que le site n'imprime que sur la page d'une fiche. releases porte les versions vidéo publiées par le
site et release_match dit s'il en a publié une : stated quand oui, none
quand non, donc une fiche marquée none ne dit rien de la vidéo sur laquelle
elle est calée. uploader vaut null sur environ deux fiches sur trois.
published_at est l'horodatage du site lu en ISO 8601 et ne porte aucun fuseau,
et published_text garde la formulation du site. rating porte deux compteurs
que le site publie, où un zéro est un chiffre qu'il a imprimé.
list_languagesListe les langues que le site catalogue, ou celles que tient une série.
Argument | Type | Requis | Ce qu'il fait |
| chaîne, 1 à 12 caractères | non | Un id de série, pour lire ce que cette série tient. |
| entier, 1 à 200 | non | Sur quelle saison mesurer. Ignoré sans |
En retour : chaque langue avec le nom que le site imprime, le site_code de
deux lettres par lequel il l'adresse, le code BCP 47 quand la correspondance
est certaine, et differs_from_iso. scope dit ce qui a été mesuré :
catalogue pour le site entier, ou season quand un id de série a été passé, et
count vaut alors les épisodes de cette saison qui tiennent la langue.
Le site dessine vingt-quatre drapeaux et adresse chaque langue par deux lettres
de son choix. Six diffèrent de l'ISO 639-1, et l'une entre en collision : le site
écrit br pour le portugais brésilien, que l'ISO attribue au breton.
Code du site | Langue | BCP 47 |
| Portugais brésilien |
|
| Grec |
|
| Tchèque |
|
| Japonais |
|
| Chinois |
|
| Ukrainien |
|
language garde le nom du site et language_code porte le tag, donc rien n'est
à déduire des deux lettres. list_subtitles accepte une langue écrite de l'une
des trois façons.
Rien n'est à régler. Toutes les variables ci-dessous sont optionnelles.
Variable | Défaut | Bornes | Ce qu'elle fait |
| absent | Placé devant l'agent du serveur, qui reste attaché. | |
| 2000 | 1500 à 60000 | Millisecondes entre deux requêtes. 1500 est un plancher. |
| 20000 | 1000 à 120000 | Délai d'une tentative. |
| 60000 | 5000 à 600000 | Délai d'une lecture, reprises comprises. |
| 3 | 0 à 8 | Tentatives après la première. |
| 900000 | 0 à 86400000 | Durée de conservation d'une page en mémoire. 0 éteint le cache. |
| 200 | 1 à 5000 | Pages gardées avant que la plus ancienne parte. |
| 8000000 | 100000 à 64000000 | Plus grande réponse lue pour une page. |
|
|
| Ce qui atteint stderr. |
Une valeur hors bornes est refusée sur stderr et le défaut s'applique.
Code | Ce que ça veut dire | Quoi faire |
| Le site a répondu et ne tient pas cette chose. | Vérifier que l'id vient d'une liste et non d'une construction. |
| Les arguments ne peuvent pas produire de requête. | Lire le message, qui nomme l'argument. |
| Le site demande à ce client de ralentir. | Attendre et redemander. La chose demandée existe toujours. |
| Une réponse est arrivée dans une forme illisible. | Le signaler avec les arguments employés. |
| La requête ne s'est pas achevée. | Réessayer. |
| Aucune réponse dans le délai. | Réessayer, ou élargir |
La couche qui lit le site est publiée seule, avec son rythme, son cache et ses codes d'erreur, sans protocole attaché.
import { TvSubtitlesClient } from "mcp-tvsubtitles/client";
const client = new TvSubtitlesClient();
const found = await client.searchShows("Smallville");
const season = await client.getSeason(found.data.rows[0].id, 0);
console.log(season.data.showName, season.data.season, season.data.episodes.length);Toute lecture rend { data, cached }, avec skipped quand des lignes ont été
écartées. Le constructeur prend { config, logger, fetchImpl }, et le plancher
d'intervalle tient quoi qu'on lui passe.
Une requête à la fois, deux secondes d'écart, élargi quand le site repousse et resserré après une série de réponses propres. Le plancher d'une seconde et demie ne peut pas être abaissé. La chaîne d'agent porte le nom du projet, sa version et l'adresse de ce dépôt, pour que le site puisse joindre une personne.
Les sous-titres sont l'œuvre de ceux qui les ont écrits et calés. Ce serveur lit
le catalogue et ne télécharge aucun fichier de sous-titres : chaque fiche porte
page_url, la page qu'un lecteur ouvre pour le télécharger. Créditer
tvsubtitles.net et lier cette page en montrant un résultat.
Ce serveur MCP n'est pas affilié à tvsubtitles.net.
Aucun compte, aucune clé, aucune télémétrie. Le seul hôte joint est
https://www.tvsubtitles.net. Les pages sont gardées quinze minutes en mémoire
et rien n'est écrit sur le disque. Les diagnostics vont sur stderr. Voir
PRIVACY.md.
npm install
npm run build:fixtures
npm test
npm run coverage
npm run checknpm run test:live fait une requête par route contre le site, et tourne chaque
nuit.
Les issues et les pull requests sont bienvenues. Voir CONTRIBUTING.md et SECURITY.md.
MIT, voir LICENSE.
Read one subtitle's record on tvsubtitles.net, from an id list_subtitles returned. The record carries the episode it belongs to, the release it was cut for, who uploaded it and when, its size and how often it was downloaded. Six of the site's ten fields are absent on some records, the author on roughly two out of three, and an absent field is null rather than blank. This server reads the catalogue: open 'page_url' to download the file.
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | A subtitle id from list_subtitles. The site answers an id it does not hold by sending the reader to its front page, which this reports as an absence. |
| Name | Required | Description |
|---|---|---|
| notes | Yes | |
| cached | Yes | |
| source | Yes | |
| subtitle | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already mark this as read-only, idempotent, and non-destructive, but the description adds meaningful behavioral detail: missing fields are null rather than blank, field absence rates vary, and invalid ids are surfaced as an absence. This materially improves the agent's ability to interpret results without contradicting 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.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is four sentences, but each one earns its place: purpose, record contents, null-behavior caveat, and download routing. It is front-loaded with the primary action and uses no filler or repetition.
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 one-parameter read tool with an output schema and strong annotations, this description is complete. It covers what the record contains, how to interpret missing fields, and how to proceed when a download is needed, leaving no critical gap for correct invocation or result handling.
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%, and the id parameter is already well documented, including the invalid-id behavior. The tool description reinforces that the id comes from list_subtitles but does not add significant semantic value beyond the schema, so the baseline of 3 is appropriate.
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 ('Read one subtitle's record') and a specific resource (tvsubtitles.net), and anchors the input to an id from list_subtitles. It is clearly distinguishable from the sibling tools, which search or list subtitles rather than read a single record.
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 clearly indicates when to use this tool: after list_subtitles has returned an id. It also gives a practical boundary by saying the server reads the catalogue and that opening 'page_url' is how to download the file. It does not explicitly name alternatives or exclusions, but the usage context is unambiguous.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List the languages tvsubtitles.net catalogues subtitles in. Called without arguments it returns the twenty-four the catalogue holds. Pass a show id from search_titles to read the languages that show actually holds, measured over one season and counted in episodes. Each entry carries the name the site prints, the two-letter code it addresses the language by, and the BCP 47 tag where that mapping is certain. Six of the site's codes differ from ISO 639-1 and one of them collides: the site writes 'br' for Brazilian Portuguese, which ISO assigns to Breton. Read this before narrowing list_subtitles by language.
| Name | Required | Description | Default |
|---|---|---|---|
| id | No | A show id from search_titles, to read what that show holds rather than what the catalogue does. Left out, the whole catalogue's languages come back. | |
| season | No | Which season to measure a show's languages over. Left out, the newest season is read. Ignored without 'id', because the catalogue has no season. |
| Name | Required | Description |
|---|---|---|
| notes | Yes | |
| scope | Yes | 'catalogue' is every language the site catalogues, whatever any show holds. 'season' is what one season of one show holds, which is what was measured when 'id' was passed. |
| cached | Yes | |
| season | Yes | The season measured, which the site states on the page it served. |
| source | Yes | |
| show_id | Yes | |
| languages | Yes | |
| show_name | Yes | |
| language_count | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already say readOnlyHint, idempotentHint, and destructiveHint=false, and the description adds meaningful behavioral detail: results are measured over one season and counted in episodes, and six site codes differ from ISO 639-1 with one collision ('br' for Brazilian Portuguese vs. Breton). This is exactly the kind of domain quirk an agent needs before filtering subtitles.
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 dense but every sentence carries distinct information: core purpose, default behavior, show-level behavior, output fields, and a critical code-collision warning. It is front-loaded with the main verb and resource, and the warning is placed where it matters before the next tool step.
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 the annotations, full schema coverage, and an output schema, the description covers everything an agent needs: both invocation modes, parameter meaning, the data returned per entry, and a domain caveat. It also connects the tool to the surrounding workflow by telling the agent to read this before narrowing list_subtitles.
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?
The input schema already describes both parameters well, so the baseline is 3. The description adds extra context beyond the schema: the no-argument call returns exactly twenty-four languages, and the show-level result is 'measured over one season and counted in episodes,' which clarifies how id and season behave in practice.
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 first sentence states exactly what the tool does: 'List the languages tvsubtitles.net catalogues subtitles in.' It names the resource (languages) and the site, and the following sentences distinguish the two modes (whole catalogue vs. a specific show), setting it apart from siblings like list_subtitles and search_titles.
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 explicit usage guidance: call without arguments for the catalogue-wide list, or pass a show id from search_titles to see what one show holds. It also tells the agent when to use this tool relative to a sibling: 'Read this before narrowing list_subtitles by language.'
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List what tvsubtitles.net holds for one series. Pass 'id' from search_titles. With 'season' alone the answer is that season's coverage, one row per episode saying which languages hold something, and 'kind' reads 'coverage'. Add 'episode' and the answer is the subtitle records themselves, with 'kind' reading 'subtitles'. Leaving 'season' out reads the newest season, and the answer says which one that was. This server reads the catalogue: each record carries the page to open to download it.
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | A show id from search_titles. The site answers an id it does not hold with a page. | |
| limit | No | Rows to render, 40 by default and 200 at most. | |
| season | No | Leave it out to read the newest season. The answer states which season it read. | |
| episode | No | Name an episode to read its subtitle records rather than the season's coverage. | |
| language | No | A language from list_languages, by its name, the site's two-letter code, or its BCP 47 tag. Left out, every language comes back. |
| Name | Required | Description |
|---|---|---|
| id | Yes | |
| kind | Yes | Which shape 'results' takes. |
| notes | Yes | |
| cached | Yes | |
| season | Yes | The season the site served, which it states on the page. |
| source | Yes | |
| episode | Yes | The episode read, or null for a season. |
| results | Yes | Coverage rows when 'kind' is 'coverage', subtitle records when it is 'subtitles'. |
| show_name | Yes | |
| result_count | Yes | |
| total_counts | Yes | What 'total_available' counted: the episodes of a season, or the records one episode held. |
| filters_applied | Yes | |
| filters_dropped | Yes | |
| total_available | Yes | |
| season_requested | Yes | The season that was asked for, or null when the newest was asked for. |
| seasons_available | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations already mark the tool read-only and idempotent; the description adds useful behavioral detail beyond that: the kind field reports coverage versus subtitles, the season default behavior, and the fact that each record carries the page to open to download it. It also notes in the schema that unknown ids return a page.
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 dense but every sentence earns its place: purpose, id source, mode semantics, default behavior, and result interpretation are all covered without fluff. The main behavioral details are front-loaded.
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 an output schema exists and all parameters are documented, the description covers the necessary call-shaping details: how to switch modes, default season behavior, and what the records contain. Nothing critical is missing for an agent to invoke this correctly.
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 coverage is 100%, so the baseline is 3, but the description goes further by explaining the relationship between season and episode, how the output changes depending on their combination, and what omitting season does. This adds real meaning beyond the individual parameter descriptions.
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 and resource: 'List what tvsubtitles.net holds for one series.' It then distinguishes its two modes, season coverage versus episode subtitles, and clearly separates it from siblings like get_subtitle and search_titles by explaining the object being listed.
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?
It gives explicit usage context: pass id from search_titles, use season alone for coverage, add episode for subtitle records, and omit season to get the newest season. It does not explicitly exclude alternatives like get_subtitle, but the mode-based guidance is clear and actionable.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search tvsubtitles.net for a television series by name. The site catalogues television only, so 'media_type' takes 'tv' or 'any' and a film is refused rather than answered with an absence. Each row carries the id the other tools take, the years the site publishes for the show, and the languages it draws a flag for. Pass 'with_counts' for how many subtitles, episodes and seasons the site holds per show: it publishes those three on its catalogue index rather than on the page a search answers with, so reading them costs one further request over a large page, and each counts the whole show. Show ids come from here and are never built by hand.
| Name | Required | Description | Default |
|---|---|---|---|
| year | No | Keep only shows whose published years cover this one. When that leaves nothing, the search is reported without it and a note says the year was set aside. | |
| limit | No | Rows to render, 20 by default and 100 at most. | |
| query | Yes | The name of a series, or part of one. The site matches on the name alone. | |
| media_type | No | 'tv' or 'any' both search the catalogue, which holds television only. 'movie' is refused, because answering it with an empty list would report an absence this site cannot establish. | |
| with_counts | No | Read each row's subtitle, episode and season counts from the site's catalogue index, which costs one further request over a page of several hundred kilobytes. Each figure counts the whole show rather than this search, each is read on its own, and a cell the index leaves empty keeps a null. |
| Name | Required | Description |
|---|---|---|
| notes | Yes | |
| query | Yes | |
| cached | Yes | True when the answer was served from memory rather than the site. |
| source | Yes | |
| results | Yes | |
| counts_scope | Yes | What the three counts count, when they were read: every season of that show together, which is a different figure from the rows this search served. Null when no counts were asked for. |
| result_count | Yes | Rows rendered here. |
| total_counts | Yes | What 'total_available' counted: the rows this one search came back with. |
| filters_applied | Yes | |
| filters_dropped | Yes | |
| total_available | Yes | Rows the site's answer held, before any were rendered. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description adds substantial behavioral detail beyond the annotations: 'movie' is refused rather than answered with an empty list, 'with_counts' triggers an extra large request and counts figures for the whole show, cells can remain null, and site coverage is television-only. It also discloses consequences such as the additional request cost. No contradiction with readOnly/openWorld/idempotent hints.
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 dense but every sentence adds distinct information: coverage limitation, media_type behavior, row contents, with_counts cost/counting semantics, and ID provenance. It is front-loaded with the core purpose and never repeats the title or schema.
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 search entry-point tool, the description covers what it returns, how parameters change behavior, cost implications, and how the returned IDs connect to other tools. An output schema also exists, so not listing return fields is acceptable; nothing an agent needs before invoking it is missing.
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 coverage is 100%, but the description still adds meaning: it explains the 'with_counts' request-cost and whole-show counting semantics, explains why 'movie' is refused despite the enum allowing it, and states that query matches on the name alone. This goes well beyond the field descriptions.
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 clear verb and resource: 'Search tvsubtitles.net for a television series by name.' It goes beyond the title by describing what each returned row contains and notes it returns the IDs that sibling tools consume, so an agent can distinguish this search tool from the get/list siblings.
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?
It gives clear context for when to call it: this is the source of show IDs for other tools ('Show ids come from here and are never built by hand'), and it states a firm exclusion — films are refused, so the tool should not be used for movie searches. It does not explicitly name sibling alternatives or say 'use list_subtitles instead,' so it stops short of a 5.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Each tool has a clear, non-overlapping role: search for shows, list available subtitle records, fetch a single record, and list languages. Even where list_subtitles and get_subtitle both concern subtitle records, their descriptions make the one-to-many relationship explicit.
All four tool names follow a consistent verb_noun pattern: search_titles, list_subtitles, list_languages, get_subtitle. The verbs correspond sensibly to the action (search vs. list vs. get), and there is no mixing of naming conventions.
Four tools is an appropriate size for a read-only TV subtitle catalogue server. Each tool covers a necessary step in the core workflow: find the show, inspect coverage, inspect subtitle records, and read a specific record.
The read-only catalogue workflow is mostly complete: search, list languages, list subtitles, and get a subtitle record cover the main needs. The main gap is that list_subtitles does not appear to accept a language filter even though list_languages says to read it 'before narrowing list_subtitles by language,' forcing client-side filtering.
YouTube transcripts, subtitles, and video metadata as structured JSON via an Apify Actor.
YouTube transcripts, search, channel/playlist listings and upload tracking for AI agents. No signup.
Extract YouTube transcripts, search what was said, and read on-screen frames with cited timestamps.
Search YouTube and read video, channel and transcript data as JSON. No Google Cloud project.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/smeet666/mcp-tvsubtitles'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
