mcp-stashbox
An MCP server for public stash-box metadata catalogues, enabling search of scenes and performers, reading records, and identifying files via fingerprints.
README
mcp-stashbox
An MCP server for the public stash-box metadata catalogues. Search scenes and performers, read one record, and identify a file from the fingerprints held for it.
These catalogues describe scenes, performers, studios and tags. They hold no media: a record names where something was published and carries nothing of it.
What it reads
| Catalogue | Address | Notes |
|---|---|---|
| StashDB | stashdb.org |
the canonical catalogue, curated by submission and review |
| ThePornDB | theporndb.net |
answers a smaller surface (see below) |
| FansDB | fansdb.cc |
independent creators |
| PMV Stash | pmvstash.org |
fan edits and music videos |
| JAVStash | javstash.org |
Every catalogue issues its own key to a registered account. A catalogue with no key is named as absent from every answer, so an answer holding rows from some of them is never read as the whole.
Setup
Each catalogue is configured through its own variable. One is enough to start.
| Variable | Catalogue |
|---|---|
STASHBOX_STASHDB_KEY |
StashDB |
STASHBOX_TPDB_KEY |
ThePornDB |
STASHBOX_FANSDB_KEY |
FansDB |
STASHBOX_PMV_KEY |
PMV Stash |
STASHBOX_JAVSTASH_KEY |
JAVStash |
A key is found on your profile page once you are registered and logged in.
{
"mcpServers": {
"stashbox": {
"command": "npx",
"args": ["-y", "mcp-stashbox"],
"env": { "STASHBOX_STASHDB_KEY": "your-key" }
}
}
}
Optional settings: SB_USER_AGENT, SB_MIN_INTERVAL_MS (default and floor
1000), SB_TIMEOUT_MS (20000), SB_MAX_RETRIES (3), SB_CACHE_TTL_MS (900000),
SB_CACHE_MAX_ENTRIES (200), SB_LOG_LEVEL (error).
Tools
| Tool | What it answers |
|---|---|
search_scenes |
scenes across every configured catalogue |
search_performers |
performers, matched on names and aliases alike |
get_scene |
one scene, with opt-in fingerprints and images |
get_performer |
one performer, with opt-in appearance, images, scenes, studios |
find_by_fingerprint |
what a file is, from the hashes held for it |
Sections are opt-in because two of them dwarf the rest: a scene's fingerprints weigh more than everything else it carries, and a performer's scenes run to thousands.
What an answer is allowed to claim
Each of these exists because breaking it produces a confident false statement.
Three fingerprint algorithms make three different claims. MD5 and OSHASH match the same file, byte for byte. PHASH matches images that resemble each other, which covers a re-encode, a crop, and a different scene from the same shoot. Every match carries its algorithm and says which kind of claim it is.
A fingerprint report that was never counted leaves the contest unknown.
One catalogue publishes how many people submitted a fingerprint and never how
many disputed it. A match from it reports contested: null, since rendering that
as false would state an agreement nobody expressed.
A scene count counts one catalogue's coverage. A settled performer record naming a career spanning decades can report zero scenes. That measures what the catalogue has indexed and states nothing about a person's work.
A folded identifier answers. When two records are joined, the older identifier still resolves and comes back as a marker naming its successor. It is a record under a new name, and its emptied fields describe the record, never the world.
A date keeps the precision it was entered with. The catalogues store dates as text, so a record carries a full day, a month or a bare year. A bare year is never printed as a day. A date shaped like a date that names none (a thirteenth month, the thirty-first of April) is read as no date at all.
A refusal is never an absence. These catalogues answer HTTP 200 with an error and a null payload, so a client reading the payload alone renders "there is no such record" where the catalogue said "I do not authorise you to ask". Errors are read first, every time.
Counts are never added and rows are never ranked across catalogues. They index overlapping corpora and publish no score in common, so rows interleave and every answer says how the order was built.
Nothing states what may be reused. A record reports the terms it carries, and silence is never read as permission.
The catalogue that answers a smaller surface
Four of these run one published open-source server. ThePornDB reimplements its interface from a source of its own. It reads records in full, and it offers no full-text search, publishes no table sorting the sites a record links to, and counts no disputes over a fingerprint. Each of those is reported on every answer holding its rows. No value is filled in from a neighbour.
As a library
The lower layer is published on its own, with the pacing, the store and the error taxonomy and no protocol attached.
import { StashboxClient } from "mcp-stashbox/client";
const client = new StashboxClient({ keys: { stashdb: process.env.STASHBOX_STASHDB_KEY } });
const read = await client.findByFingerprint({
fingerprints: [{ hash: "…", algorithm: "PHASH" }],
});
The floor on pacing holds through this entry point too: a consumer cannot ask these catalogues for more than the server would.
What this owes the catalogues
They are run by their communities and paid for by donations, and none of them
publishes a rate limit. One request at a time per catalogue, spaced by at least a
second, widening when a catalogue pushes back. The User-Agent always carries
this project's identifier and an address where a person can be reached.
Licence
MIT. See LICENSE.
mcp-stashbox (français)
Un serveur MCP pour les catalogues de métadonnées stash-box publics. Chercher des scènes et des personnes, lire une fiche, et identifier un fichier à partir des empreintes qu'on en détient.
Ces catalogues décrivent des scènes, des personnes, des studios et des tags. Ils ne détiennent aucun média : une fiche nomme le lieu de publication et n'en porte rien.
Ce qu'il lit
StashDB (stashdb.org), ThePornDB (theporndb.net), FansDB (fansdb.cc),
PMV Stash (pmvstash.org) et JAVStash (javstash.org). Chaque catalogue délivre
sa propre clé à un compte inscrit. Un catalogue sans clé est nommé comme absent
de chaque réponse, jamais retiré en silence.
Configuration
STASHBOX_STASHDB_KEY, STASHBOX_TPDB_KEY, STASHBOX_FANSDB_KEY,
STASHBOX_PMV_KEY, STASHBOX_JAVSTASH_KEY. Une seule suffit pour commencer. La
clé se lit sur la page de profil une fois connecté.
Réglages facultatifs : SB_USER_AGENT, SB_MIN_INTERVAL_MS (défaut et plancher
1000), SB_TIMEOUT_MS (20000), SB_MAX_RETRIES (3), SB_CACHE_TTL_MS (900000),
SB_CACHE_MAX_ENTRIES (200), SB_LOG_LEVEL (error).
Les outils
search_scenes, search_performers, get_scene, get_performer et
find_by_fingerprint. Les sections sont facultatives parce que deux d'entre
elles écrasent les autres : les empreintes d'une scène pèsent plus que tout ce
qu'elle porte, et les scènes d'une personne se comptent par milliers.
Ce qu'une réponse a le droit d'affirmer
Trois algorithmes d'empreinte, trois affirmations différentes. MD5 et OSHASH désignent le même fichier, octet pour octet. PHASH désigne des images qui se ressemblent, ce qui couvre un ré-encodage, un recadrage et une autre scène du même tournage.
Une contestation jamais comptée laisse la contestation inconnue. Un
catalogue publie le nombre de soumissions d'une empreinte sans jamais publier le
nombre de contestations. Une correspondance venue de lui rend contested: null.
Un compteur de scènes mesure la couverture d'un catalogue. Une fiche établie annonçant une carrière de plusieurs décennies peut en porter zéro.
Un identifiant fusionné répond. Il revient sous forme de marqueur nommant son successeur : une fiche sous un autre nom plutôt qu'une absence.
Une date garde la précision avec laquelle elle a été saisie. Une année seule n'est jamais rendue comme un jour, et une date qui n'en nomme aucun, un treizième mois ou le 31 avril, n'est pas une date.
Un refus n'est jamais une absence. Ces catalogues répondent 200 avec une erreur et une charge nulle, donc les erreurs sont lues avant la charge.
Les compteurs ne s'additionnent pas et les lignes ne se classent pas entre catalogues. Elles s'entrelacent, et chaque réponse dit comment l'ordre a été construit.
Ce qu'on doit à ces catalogues
Ils sont communautaires et financés par dons, et aucun ne publie de limite. Une
requête à la fois par catalogue, espacée d'au moins une seconde, élargie quand un
catalogue le demande. Le User-Agent porte toujours l'identifiant du projet et
une adresse où joindre une personne.
Licence
MIT. Voir LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.