de en es ja ko ru zh

Pexafy MCP Server

Name: Pexafy MCP Server
Author: Pexafy

Official

by Pexafy

Overview Schema Related Servers Score Discussions

Python

Hybrid

pexafy-mcp

License: MIT

Stockfotosuche für KI-Assistenten. Ein MCP-Server, der Claude, ChatGPT oder jedem anderen MCP-Client die Suche in einer Bibliothek lizenzfreier Bilder erlaubt – durch die Beschreibung einer Szene in natürlicher Sprache, anhand eines Beispielbilds oder „mehr davon“ – und die Ergebnisse als Thumbnail-Raster direkt in der Konversation anzeigt.

Remote-MCP, OAuth, kein API-Schlüssel zum Einfügen, 3 Tools, Bilder werden direkt gerendert.

Das Pexafy-Ergebnisraster, inline in einer Claude-Konversation gerendert

Nutzung (nichts zu installieren)

Ein gehosteter Server läuft unter:

https://mcp.pexafy.com/mcp

Er spricht Streamable HTTP und authentifiziert sich mit OAuth 2.1 – du meldest dich in einem Browserfenster bei Pexafy an und der Connector erhält die eigenen Zugangsdaten. Es gibt keinen API-Schlüssel, der erzeugt, in eine JSON-Datei eingefügt oder später erneuert werden müsste.

Claude (Web und Desktop)

Öffne Einstellungen → Connectors (bei Team/Enterprise fügt ein Besitzer ihn einmal unter Organisationseinstellungen → Connectors hinzu).
Klicke auf Benutzerdefinierten Connector hinzufügen.
Füge https://mcp.pexafy.com/mcp ein und bestätige.
Melde dich in dem sich öffnenden Fenster bei Pexafy an. Fertig – du kannst Claude jetzt nach einem Foto fragen.

Claude Code

claude mcp add --transport http pexafy https://mcp.pexafy.com/mcp

Jeder andere MCP-Client

Richte ihn mit dem streamable-http-Transport auf dieselbe URL aus. Clients, die kein OAuth implementieren, können sich stattdessen mit einem Pexafy-API-Schlüssel erstellt werden, der als Authorization: Bearer <key> oder x-api-key: <key>, erhalten – einen bekommst du im Dashboard.

Erreichbarkeit: GET /health (öffentlich, keine Authentifizierung).

Außerdem gelistet im offiziellen MCP-Registry als com.pexafy/pexafy-mcp und auf Smithery – dort ist eine gehostete Gateway-URL für Clients verfügbar, die das bevorzugen.

Was es kostet

Der kostenlos-Tarif umfasst 5.000 Suchen pro Monat mit einem Connector – genug für den regelmäßigen Gebrauch, ohne dass eine Kreditkarte erforderlich ist. Höhere Stufen findest du auf der Preisseite. Wenn du ein Limit erreichst, teilt dir dies dem Assistenten im Chat mit, statt in einem nur schwer verständlichen Fehler zu scheitern.

Related MCP server: brave-image-mcp

Tools

Drei schreibgeschützte Tools – ohne Schreibzugriff, ohne Kontomutation.

`search_photos` – semantische Textsuche

Beschreibe die Szene im einem vollständigen Satz; Pexafy arbeitet semantisch, daher sind Sätze besser als Schlüsselwörter. Alle Parameter sind optional, aber gib entweder q oder mindestens einen Filter an.

Parameter	Typ	Hinweise
`q`	string	Die Szene in natürlicher Sprache. Maximal 500 Zeichen.
`color_name`	string	Eine der: red, orange, yellow, green, blue, purple, pink, brown, black, white, gray, teal, beige, gold, navy. Schließt `color_hex` aus.
`color_hex`	string	z. B. `#1E90FF`. Schließt `color_name` aus.
`color_tolerance`	integer	0 (exakt) bis 255 (ungenau). Standard: 20. Nur in Verbindung mit `color_hex`.
`orientation`	string[]	`landscape`, `portrait`, `square`.
`src`	string[]	Unsplash, Pexels, Pixabay, Kaboompics, Burst, StockSnap, Picjumbo, Skitterphoto, NegativeSpace.
`license_type`	string[]	`free`, `cc0`.
`photographer`	string	Exakter Benutzername.
`after_date`	string	`YYYY-MM-DD`. Veröffentlicht am oder nach diesem Tag.
`cursor`	string	`pagination.next_cursor` aus einer früheren Antwort. (der)

`search_photos_by_image` – visuelle Suche anhand eines Beispielbilds

Findet Fotos, die einem Referenzbild ähneln, optional um Textangaben ergänzt („so, aber nachts“).

Parameter	Typ	Hinweise
`image_url`	string	Öffentliche http(s)-URL des Referenzbildes.
`image_file`	object	Wird von Hosts mit Upload-Support vollautomatisch ausgefüllt (z. B. ChatGPT).
`image_base64`	string	Text-Base64-Bytes für programmatische Clients.
`q`	string	Text, der mit dem Bild kombiniert wird („mit erhobenen Händen“).
`text_alpha`	number	Gewicht von `q` gegenüber dem Bild.
`orientation`, `source`, `color_name`, `license_type`, `photographer`, `after_date`	string	Gleiche Filter wie oben.
`cursor`	string	Pagination-Token.

Einer von image_url, image_file oder image_base64 ist erforderlich. Bilder werden serverseitig abgerufen; maximal 20 MB.

"das findest du herum" – mehr davon

Parameter	Typ	Hinweise
`photo_id`	string	Erforderlich. Die UUID eines Fotos aus einem früheren Ergebnis.
`cursor`	string	Pagination-Token.

Was kommt zurück

Jedes Foto hat seine ID, URLs in mehreren Größen, Maße, Zahl, dominante Farbe, Ausrichtung, Quelle, Lizenz, Fotografin/Fotograf und einen attribution-String, der als Anerkennung anzeigt wird – genug, damit der Assistent über die Ergebnisse nachdenken kann, statt sie nur aufzuzählen.

Die Ergebnisse sind #1, #2, … nummeriert, damit kannst du dich so auf ein Foto ziehen, wie du es in einem Gespräch tun würdest. Keine IDs zum Kopieren:

Mehr Fotos wie #1 anfragen und Darstellung des Assistenten über die neue Menge

In Clients, die MCP-Apps unterstütz, öffnen Sie bei Klick auf die Miniaturansicht ein Detail-Panel mit den gesamten Metadaten – ohne zusätzlichen nächsten Aufruf; alles ist bereits im Tool-Ergebnis vorhanden:

Das Detail-Panel: Fotograf, Quelle, Auflösung, Lizenz, dominante Farbe, Ausrichtung und Beschreibung

Selbst hosten

Das musst du nicht – der gehostete Server ist oben, so ist vorgesehen. Aber der Server ist ein leichter, einfacher Client für die Pexafy-API, du kannst also einen eigenen mit dem eigenen Schlüssel betreiben.

Erfordert Python 3.12+.

git clone https://github.com/Pexafy/pexafy-mcp.git && cd pexafy-mcp
./run.sh setup        # venv + editable install + seed .env
# edit .env — set PEXAFY_API_KEY
./run.sh dev          # stdio, for Claude Desktop / Claude Code

Mit dem installierten Konsolen-Script (pip install .):

pexafy-mcp                             # stdio (default)
PEXAFY_MCP_TRANSPORT=http pexafy-mcp   # remote Streamable HTTP

Claude-Desktop / Claude Code, über stdio:

{
  "mcpServers": {
    "pexafy": {
      "command": "pexafy-mcp",
      "env": { "PEXAFY_API_KEY": "pexafy_api_…" }
    }
  }
}

Docker, über HTTP — siehe docker-compose.example.yml:

docker compose -f docker-compose.example.yml up -d
curl localhost:8765/health

Das Image selbst hat standardmäßig stdio – der Transport, den ein MCP-Client zum Steuern eines Containers verwendet – deshalb funktioniert es auch direkt:

docker run -i --rm pexafy-mcp

Das beantwortet initialize und tools/list ohne API-Key und ohne Netzwerk, denn die Tools kommen aus dem mitgelieferten OpenAPI-Snapshot. Ein Schlüssel wird nur für die Suche benötigt. Über HTTP zu bedienen, ist eine Frage des gesetzten Transports, und genau das tun die beiden Compose-Dateien.

Konfiguration

Jede Einstellung ist eine Umgebungsvariable, und jede ist optional: Ohne gesetzte Optionen startet pexafy-mcp über stdio und beantwortet initialize und tools/list offline. Zwei sind wissenswert.

Variable	Standardwert	Zweck
`PEXAFY_MCP_TRANSPORT`	`stdio`	`stdio` für einen lokalen Client, `http` für den Remote-Betrieb
`PEXAFY_API_BASE_URL`	`http://localhost:8000`	Pexafy-API-Wurzel – punkt auf `https://api.pexafy.com` oder deine eigene Bereitstellung

Der Rest gehört eher zu einem Deployment als zu jemand, der den Container ausführt, und liegt in .env.example: ein Fallback-PEXAFY_API_KEY für stdio-Verwendung, wenn der Client keinen eigenen Schlüssel schickt, PEXAFY_THUMB_BASE_URL und PEXAFY_THUMB_HMAC_SECRET um die Thumbnails hinter dem Inline-Raster zu signieren, sowie PEXAFY_OAUTH_* zusammen mit MCP_RESOLVE_SECRET, um den HTTP-Transport als OAuth-Ressource-Server zu betreiben. Keine davon brauchts es, um den Server zu starten.

Wie es funktioniert

src/pexafy_mcp/
├── server.py     # entry point: builds the server, wires hooks, custom tools, /health
├── tooling.py    # tunes the OpenAPI-derived tools for an LLM (descriptions, value sets)
├── widget.py     # MCP Apps UI resource — the inline result grid (self-contained HTML)
├── previews.py   # signs the thumbnail URLs injected into each result
├── limits.py     # turns plan-limit (429) responses into in-chat upgrade nudges
├── auth.py       # per-user auth: OAuth Resource Server or forwarded API key
└── assets/       # vendored, shipped with the package:
    ├── openapi.json          # OpenAPI snapshot the tools are generated from
    ├── facets.json           # evolving source/license value sets
    └── ext_apps_bundle.js    # @modelcontextprotocol/ext-apps SDK (inlined in the widget)

Die Tools werden erzeugt aus der Pexafy-OpenAPI-Spezifikation über FastMCP.from_openapi(), damit die API die einzige Quelle bleibt; tooling.py formt sie anschließend für das Sprachmodell – verkleinert die Oberfläche auf den Kernbereich Suche, entfernt die Parameter, durch die sich Modell irreführend werden, und integriert die feststehenden Wertemengen, so dass nie eine Facetten-Abfrage nötig ist.
build_server() setzt alles zusammen. Ein Import des Pakets hat keine Seiteneffekte und macht keine Netzwerk-I/O: es liest das mitgelieferte assets/openapi.json und assets/facets.json. Das Viertelprepare.sh` regeneriert diese.
search_photos_by_image ist handschrift: Ein Chat-Assistent kann keine Binärdatei an ein MCP-Tool übergeben, also akzeptiert das Tool eine Bild-URL und lädt es serverseitig ab.
Das Inline-Raster ist eine MCP-Apps-UI-Ressource. Der ext-apps-Client ist mitgebündelt und eingebettet, weil die iframe des Host-Sandbox keine externen Skripte zur Laufzeit abrufen kann.

Entwicklung

./run.sh test         # offline test suite (pytest)
./run.sh inspect      # MCP Inspector
./prepare.sh          # maintainers: regenerate the vendored assets/

Beiträge willkommen – siehe CONTRIBUTING.md.

Lizenz

MIT – siehe LICENSE.

Das Paket überträgt außerdem Drittanbieter-Ressourcen (die Inter-Schriftart, den @modelcontextprotocol/ext-apps-Browserbundles und die darin enthaltenen Bibliotheken), jeweils unter eigener Lizenz – siehe THIRD _PARTY _NOTICES.md.

Available Tools

3 tools

get_similar_photosFind similar photosA

Read-onlyIdempotent

Inspect

Use this tool when the user says 'find something similar', 'show me more like this', or 'I need a visually consistent set'. Requires a photo_id obtained from a previous search result. A person normally refers to a photo by the rank shown on the result grid rather than by its identifier; each search result carries both, in the same object. Each result carries: rank, its position on this page (1, 2, 3, …), which is also the number drawn on the inline grid and the handle a person naturally uses to refer to one photo among several; photo_id, the identifier the similar-photos tool takes, present in the same object as the rank; attribution, the credit line to display with the photo; and urls, the image at several sizes, urls.regular being the one to link to. Inline thumbnails are attached to this tool's result as an MCP App resource. Some clients, claude.ai on the web among them, render that resource only inside an expandable tool panel rather than in the reply itself; where it is not rendered, the photos remain reachable through their URLs.

ParametersJSON Schema

Name	Required	Description	Default
`cursor`	No	Token to fetch the next page. Take `pagination.next_cursor` from a previous response and pass it back here. See the Pagination guide.
`photo_id`	Yes	The reference photo's unique Pexafy identifier (a UUID).

Output Schema

ParametersJSON Schema

Name	Required	Description
`data`	No
`meta`	No
`error`	No
`success`	No
`pagination`	No

TDQS

A4.3/5.0

Behavior4/5

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

Annotations already indicate readOnlyHint, openWorldHint, idempotentHint, and non-destructive behavior, so the safety profile is covered. The description adds useful behavioral context: the result includes inline thumbnails as an MCP App resource, and warns that some clients (like claude.ai on web) render them only in an expandable panel, with URLs still available. This goes beyond annotations and is valuable for the agent.

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

Conciseness4/5

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

The description is relatively lengthy but every paragraph serves a purpose: it explains when to use, prerequisites, how users refer to photos (critical for agent understanding), and display behavior. It is front-loaded with the primary usage trigger and then provides necessary details. Slightly verbose but justified by the need to explain the rank vs. photo_id distinction.

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?

The tool has an output schema, so return values are covered there. The description compensates for the complexity of the tool by explaining the relationship between rank and photo_id, which is not obvious from the schema. It also addresses pagination and resource rendering behavior. Given the moderate complexity and presence of output schema, this is adequately complete, though more details on what 'similar' entails could be added.

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?

Schema coverage is 100%, with both parameters (photo_id and cursor) described in the schema. The description reinforces the use of photo_id (requires it from a prior search) and explains the cursor's role (pass pagination.next_cursor), but adds minimal additional semantics beyond the schema. Baseline 3 is appropriate given 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 clearly states it finds similar photos based on a photo_id, distinct from sibling search tools by focusing on similarity rather than keywords or image upload. It explicitly ties to user phrases like 'find something similar', making its purpose actionable.

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?

It explicitly explains when to use the tool ('when the user says...'), specifies the prerequisite (photo_id from a previous search), and details how a person refers to photos (by rank) versus the identifier, which prevents misuse. It also clarifies how to use the cursor for pagination.

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

search_photosSearch photos by descriptionA

Read-onlyIdempotent

Inspect

Use this tool whenever the user needs an image, photo, or visual — for a presentation, blog, website, social-media post, mood board, or any creative project. Pexafy is a SEMANTIC search engine: describe the scene in full natural-language sentences, not keywords. Rich descriptions return far better results than tag-like queries. Good queries: 'a melancholy portrait of an old person sitting under a soft light'; 'two people sharing a bench in comfortable silence'; 'the last sunlight of the day hitting a dusty windowsill'; 'a child discovering snow for the first time'. Prefer this tool over search_photos_by_image when the user describes what they want in words. BUT if they want photos LIKE a specific image that has a URL — a photo from a previous result, or a public URL they gave — use search_photos_by_image instead (pass that URL, plus a q for any change like 'but with hands raised'). Only use THIS text tool for a reference image with NO URL (a file pasted/uploaded in the chat): describe what you see in rich detail — Pexafy is semantic, so a good description finds visually similar photos. Each result carries: rank, its position on this page (1, 2, 3, …), which is also the number drawn on the inline grid and the handle a person naturally uses to refer to one photo among several; photo_id, the identifier the similar-photos tool takes, present in the same object as the rank; attribution, the credit line to display with the photo; and urls, the image at several sizes, urls.regular being the one to link to. Inline thumbnails are attached to this tool's result as an MCP App resource. Some clients, claude.ai on the web among them, render that resource only inside an expandable tool panel rather than in the reply itself; where it is not rendered, the photos remain reachable through their URLs.

ParametersJSON Schema

Name	Required	Description
`q`	No	Your search query as a full natural-language sentence describing the scene you want — Pexafy is semantic, so sentences beat keywords. Up to 500 characters. Optional if you provide at least one filter instead. Example: 'an old man sitting at a café table he has visited every morning for thirty years'.
`cursor`	No	Token used to fetch the next page. Take the `pagination.next_cursor` value from a previous response and pass it back here. See the [Pagination](/pagination) guide.
`source`	No	Keep only photos from these providers: Unsplash, Pexels, Pixabay, Kaboompics, Burst, StockSnap, Picjumbo, Skitterphoto, NegativeSpace. Repeat the parameter to pass several.
`color_hex`	No	Keep only photos close to this hex color (e.g. `#1E90FF`). Cannot be combined with `color_name`. Use `color_tolerance` to widen or tighten the match.
`after_date`	No	Only return photos published on or after this date, formatted `YYYY-MM-DD`.
`color_name`	No	Keep only photos whose dominant color matches one of: red, orange, yellow, green, blue, purple, pink, brown, black, white, gray, teal, beige, gold, navy. Cannot be combined with color_hex.
`orientation`	No	Keep only photos with these shapes: landscape, portrait, square. Repeat the parameter to pass several.
`license_type`	No	Keep only photos with these license types: free, cc0. 'free' means the photo can be used freely and attribution is appreciated. Repeat the parameter to pass several.
`photographer`	No	Only return photos from this photographer's username. Use `GET /api/v1/facets/photographers/suggest` to find usernames.
`color_tolerance`	No	How far a photo's color may be from `color_hex` and still match, from `0` (exact match) to `255` (very loose). Defaults to `20`. Only applies when `color_hex` is set.

Output Schema

ParametersJSON Schema

Name	Required	Description
`data`	No
`meta`	No
`error`	No
`success`	No
`pagination`	No

TDQS

A4.8/5.0

Behavior5/5

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

The annotations already declare a safe, read-only, idempotent operation, so the description's job is to add context beyond that. It does: semantic search behavior, result-field semantics (rank, photo_id, attribution, urls), the inline-thumbnail MCP resource, and the rendering caveat on claude.ai. No contradiction with annotations.

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

Conciseness4/5

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

The description is long, but it is front-loaded with the primary use case and every section earns its place: query style, examples, sibling distinction, result fields, and rendering behavior. A few example queries could be trimmed without losing meaning, which keeps it from a 5.

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 tool with 10 optional parameters, an output schema, and two siblings, the description is complete: it explains semantic querying, when to use each sibling, what each result field means, and how the inline resource may render. The output schema covers return values, so the description correctly focuses on selection and invocation behavior.

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 and the schema already documents every parameter. The description adds real value by teaching the core q semantics, showing strong example queries, and explaining how photo_id connects to the similar-photos sibling, but it does not need to repeat the schema.

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 opens with a specific use case ('user needs an image, photo, or visual') and names the resource being searched. It clearly differentiates this text-query tool from search_photos_by_image, which is the main sibling it could be confused with.

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?

It gives explicit when-to-use guidance ('Prefer this tool over search_photos_by_image when the user describes what they want in words') and names the alternative with the exact input it needs. It also handles the edge case of a reference image with no URL, telling the agent to describe it in rich detail instead.

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

search_photos_by_imageSearch photos by example imageA

Read-onlyIdempotent

Inspect

Find visually similar stock photos from an EXAMPLE IMAGE, optionally TWEAKED with words. This is the right tool for 'find photos LIKE THIS but ' (e.g. 'like this but with their hands raised', 'the same scene but at night'). Give the reference image one of three ways: (1) image_url — a public http(s) link: a photo from a PREVIOUS search result (reuse its image_url/urls.regular), or any public URL the user provides; (2) image_file — auto-filled by the host when the user UPLOADS an image (e.g. ChatGPT) — it is populated by the host, not by the caller; (3) image_base64 — raw base64 image bytes, for a programmatic client that already holds the file. A chat assistant has no access to the exact bytes of an image it was shown, so image_base64 is not available to it. Put any change in q; raise text_alpha to weight the text more. If the reference image has no URL and the host did not auto-provide image_file (e.g. a file pasted into a chat that can't be forwarded), you cannot send it — describe what you see and use search_photos instead. Every result carries an attribution you show.

ParametersJSON Schema

Name	Required	Description
`q`	No	Your search query as a full natural-language sentence describing the scene you want — Pexafy is semantic, so sentences beat keywords. Up to 500 characters. Optional if you provide at least one filter instead. Example: 'an old man sitting at a café table he has visited every morning for thirty years'.
`cursor`	No	Token to fetch the next page. Take `pagination.next_cursor` from a previous response and pass it back here — no need to re-upload the image. See the Pagination guide.
`source`	No	Keep only photos from these providers: Unsplash, Pexels, Pixabay, Kaboompics, Burst, StockSnap, Picjumbo, Skitterphoto, NegativeSpace. Repeat the parameter to pass several.
`image_url`	No	Public http(s) URL of the reference image. Reuse the `image_url` of a photo from a previous search result, or any public URL the user provides.
`after_date`	No	Only return photos published on or after this date, formatted YYYY-MM-DD.
`color_name`	No	Keep only photos whose dominant color matches one of: red, orange, yellow, green, blue, purple, pink, brown, black, white, gray, teal, beige, gold, navy. Cannot be combined with color_hex.
`image_file`	No	Filled in by the host when the user uploads an image, not by the caller. Carries the upload's `download_url` and `file_id`.
`text_alpha`	No	Balance between your text and the image when both are provided, from `0` to `10`. `0` ignores the text (pure visual search), `1.7` (the default) is balanced, and higher values give your words more weight. Has no effect without `q`.
`orientation`	No	Keep only photos with these shapes: landscape, portrait, square. Repeat the parameter to pass several.
`image_base64`	No	The reference image as base64 bytes, optionally as a `data:` URL. For a client that already holds the bytes; prefer `image_url` when a link exists.
`license_type`	No	Keep only photos with these license types: free, cc0. 'free' means the photo can be used freely and attribution is appreciated. Repeat the parameter to pass several.
`photographer`	No	Only return photos from this photographer's exact username.

Output Schema

ParametersJSON Schema

Name	Required	Description
`data`	No
`meta`	No
`error`	No
`success`	No
`pagination`	No

TDQS

A4.6/5.0

Behavior4/5

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

Annotations already declare readOnly and idempotent, so the description doesn't need to repeat that. It adds meaningful context beyond annotations: the image_file is host-populated rather than caller-set, image_base64 is unavailable to chat assistants, and every result carries an attribution to display. No contradictions.

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

Conciseness4/5

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

The description is long but every section earns its place—it covers usage, input methods, edge cases, and attribution. The numbered list of image-providing options is clear and well-structured. It could be slightly trimmed, but the density is justified by the tool's complexity.

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 12 parameters, 100% schema coverage, and an output schema, the description provides all necessary behavioral context: how to provide the reference image, the host-filling behavior of image_file, the text weighting mechanism, and the fallback to search_photos. It also mentions the attribution requirement from results, which is not in the schema.

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 documents each parameter. The description adds practical nuance beyond the schema, such as how text_alpha weights text against image, and the guidance to put any modification in q. This exceeds the baseline for fully covered schemas.

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 finds visually similar stock photos from an example image, optionally tweaked with words. It explicitly distinguishes from siblings by providing a usage scenario ('find photos LIKE THIS but <change>') and names the alternative (search_photos) when the image can't be sent.

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?

Provides explicit when-to-use guidance with examples and a concrete fallback: when the image has no URL and no auto-provided file, use search_photos instead. It also explains the three ways to supply the reference image and which is appropriate for different clients.

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. Dates show when Glama detected each change.

1 tool updatev0.4.9Aug 22, 2026

Changedsearch_photos_by_image19 fields changed

addedInput schema / properties / after_date / description

Added value: +"Only return photos published on or after this date, formatted YYYY-MM-DD."

addedInput schema / properties / color_name / description

Added value: +"Keep only photos whose dominant color matches one of: red, orange, yellow, green, blue, purple, pink, brown, black, white, gray, teal, beige, gold, navy. Cannot be combined with color_hex."

addedInput schema / properties / cursor / description

Added value: +"Token to fetch the next page. Take `pagination.next_cursor` from a previous response and pass it back here — no need to re-upload the image. See the Pagination guide."

addedInput schema / properties / image_base64 / description

Added value: +"The reference image as base64 bytes, optionally as a `data:` URL. For a client that already holds the bytes; prefer `image_url` when a link exists."

addedInput schema / properties / image_file / additionalProperties
```
Added value: +false
```

removedInput schema / properties / image_file / anyOf

Removed value: -[
-  {
-    "additionalProperties": true,
-    "type": "object"
-  },
-  {
-    "type": "null"
-  }
-]

removedInput schema / properties / image_file / default
```
Removed value: -null
```

addedInput schema / properties / image_file / description

Added value: +"Filled in by the host when the user uploads an image, not by the caller. Carries the upload's `download_url` and `file_id`."

addedInput schema / properties / image_file / properties

Added value: +{
+  "download_url": {
+    "type": "string"
+  },
+  "file_id": {
+    "type": "string"
+  },
+  "file_name": {
+    "type": "string"
+  },
+  "mime_type": {
+    "type": "string"
+  }
+}

addedInput schema / properties / image_file / required
```
Added value: +[
+  "download_url",
+  "file_id"
+]
```
addedInput schema / properties / image_file / type
```
Added value: +"object"
```

addedInput schema / properties / image_url / description

Added value: +"Public http(s) URL of the reference image. Reuse the `image_url` of a photo from a previous search result, or any public URL the user provides."

addedInput schema / properties / license_type / description

Added value: +"Keep only photos with these license types: free, cc0. 'free' means the photo can be used freely and attribution is appreciated. Repeat the parameter to pass several."

addedInput schema / properties / orientation / description

Added value: +"Keep only photos with these shapes: landscape, portrait, square. Repeat the parameter to pass several."

addedInput schema / properties / photographer / description

Added value: +"Only return photos from this photographer's exact username."

addedInput schema / properties / q / description

Added value: +"Your search query as a full natural-language sentence describing the scene you want — Pexafy is semantic, so sentences beat keywords. Up to 500 characters. Optional if you provide at least one filter instead. Example: 'an old man sitting at a café table he has visited every morning for thirty years'."

addedInput schema / properties / source / description

Added value: +"Keep only photos from these providers: Unsplash, Pexels, Pixabay, Kaboompics, Burst, StockSnap, Picjumbo, Skitterphoto, NegativeSpace. Repeat the parameter to pass several."

addedInput schema / properties / text_alpha / description

Added value: +"Balance between your text and the image when both are provided, from `0` to `10`. `0` ignores the text (pure visual search), `1.7` (the default) is balanced, and higher values give your words more weight. Has no effect without `q`."

changedOutput schema / (root)

Previous value: -nullNew value: +{
+  "properties": {
+    "data": {
+      "items": {
+        "description": "A photo result. Fields returned can be narrowed with the `fields` parameter and may depend on your plan.",
+        "properties": {
+          "alt_description": {
+            "description": "Accessibility-friendly text.",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "attribution": {
+            "description": "Ready-to-display credit for the photographer/source.",
+            "properties": {
+              "html": {
+                "description": "HTML attribution snippet.",
+                "type": "string"
+              },
+              "plain": {
+                "description": "Plain-text attribution.",
+                "type": "string"
+              }
+            },
+            "type": "object"
+          },
+          "blur_hash": {
+            "description": "BlurHash placeholder string.",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "color_hex": {
+            "description": "Dominant color hex code.",
+            "type": "string"
+          },
+          "color_name": {
+            "description": "Dominant color name.",
+            "type": "string"
+          },
+          "description": {
+            "description": "AI-generated caption.",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "height": {
+            "type": [
+              "integer",
+              "null"
+            ]
+          },
+          "image_url": {
+            "description": "Canonical source image URL.",
+            "format": "uri",
+            "type": "string"
+          },
+          "license_type": {
+            "description": "License type (e.g. `free`).",
+            "type": "string"
+          },
+          "orientation": {
+            "enum": [
+              "landscape",
+              "portrait",
+              "square"
+            ],
+            "type": "string"
+          },
+          "photo_id": {
+            "description": "Unique Pexafy identifier (UUID).",
+            "type": "string"
+          },
+          "photographer_full_name": {
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "photographer_url": {
+            "format": "uri",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "photographer_username": {
+            "type": "string"
+          },
+          "relevance_score": {
+            "description": "Match score 0–1 (higher is better). Only on search results.",
+            "type": [
+              "number",
+              "null"
+            ]
+          },
+          "source": {
+            "description": "Provider (e.g. `Pexels`, `Unsplash`, `Pixabay`).",
+            "type": "string"
+          },
+          "source_description": {
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "source_image_url": {
+            "description": "URL of the photo's page on the provider.",
+            "format": "uri",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "uploaded_on": {
+            "description": "Publication date (YYYY-MM-DD).",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
+          "urls": {
+            "description": "Ready-to-use image links in five sizes.",
+            "properties": {
+              "full": {
+                "format": "uri",
+                "type": "string"
+              },
+              "large": {
+                "format": "uri",
+                "type": "string"
+              },
+              "regular": {
+                "format": "uri",
+                "type": "string"
+              },
+              "small": {
+                "format": "uri",
+                "type": "string"
+              },
+              "thumb": {
+                "format": "uri",
+                "type": "string"
+              }
+            },
+            "type": "object"
+          },
+          "width": {
+            "type": [
+              "integer",
+              "null"
+            ]
+          }
+        },
+        "type": "object"
+      },
+      "type": "array"
+    },
+    "error": {
+      "anyOf": [
+        {
+          "properties": {
+            "code": {
+              "description": "Machine-readable error code (e.g. `MISSING_PARAMS`, `PHOTO_NOT_FOUND`).",
+              "type": "string"
+            },
+            "message": {
+              "description": "Human-readable error message.",
+              "type": "string"
+            },
+            "request_id": {
+              "type": "string"
+            }
+          },
+          "required": [
+            "code",
+            "message"
+          ],
+          "type": "object"
+        },
+        {
+          "type": "null"
+        }
+      ]
+    },
+    "meta": {
+      "properties": {
+        "request_id": {
+          "description": "Unique id for this request (quote it in support tickets).",
+          "type": "string"
+        },
+        "took_ms": {
+          "description": "Server processing time in milliseconds.",
+          "type": "number"
+        }
+      },
+      "type": "object"
+    },
+    "pagination": {
+      "anyOf": [
+        {
+          "properties": {
+            "has_more": {
+              "description": "Whether another page exists.",
+              "type": "boolean"
+            },
+            "next_cursor": {
+              "description": "Pass back as `cursor` for the next page; `null` when `has_more` is false.",
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "per_page": {
+              "description": "Number of items per page.",
+              "type": "integer"
+            }
+          },
+          "type": "object"
+        },
+        {
+          "type": "null"
+        }
+      ]
+    },
+    "success": {
+      "type": "boolean"
+    }
+  },
+  "type": "object",
+  "x-fastmcp-top-level-schema": "PhotoListResponse"
+}

2 tool updatesv0.4.0Aug 19, 2026
- Addedget_similar_photos
- Removedphoto_similar
3 tool updatesv0.2.0Aug 17, 2026
- First observedphoto_similar
- First observedsearch_photos
- First observedsearch_photos_by_image

Install Server

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

TDQS

A4.4/5.0

Disambiguation4/5

Each tool has a clearly documented input type (text query vs. image/file vs. previous photo_id), and the descriptions are explicit about which phrase or condition triggers each tool. However, search_photos_by_image and get_similar_photos both produce visually similar photos, and their boundary (one tweaks by text, the other just fetches similar) could occasionally mislead an agent even with the detailed guidance.

Naming Consistency4/5

All names follow a snake_case verb_noun pattern (search_photos, search_photos_by_image, get_similar_photos), and the shared 'search_photos' prefix on two tools is helpful. The slight deviation is 'get' in get_similar_photos versus 'search' elsewhere for the same core concept, but the pattern is otherwise uniform and predictable.

Tool Count4/5

Three tools is a lean but sensible footprint for a dedicated photo-search server, covering the natural query modalities (text, image, similar-by-id). While each tool does earn its place, the set feels slightly minimal—no dedicated tool for fetching individual photo details, but results already carry URLs and attribution, so it works.

Completeness4/5

The core workflow is complete: text query → results → similar-by-photo_id, and image query → results with tweakable text, covering the main stock-photo search use cases with no dead ends. Minor gaps exist (no downloadable/collections/curated feed support, no orientation/filter parameters), but agents can work around these with richer natural-language calls to search_photos.

Maintenance

ActivityMaintained

ResponsivenessNo issues

Resources

Need Help?

Related Servers

Related MCP Connectors

mcp-qwen-imageOAuth
io.github.AceDataCloud
MCP server for Qwen Image 3 AI image generation
AI & Machine Learning Image & Video Processing
1
mcp-wanOAuth
io.github.AceDataCloud
MCP server for Wan AI video generation
AI & Machine Learning Image & Video Processing
21
MidjourneyOAuth
io.github.AceDataCloud
MCP server for Midjourney AI image generation and editing
Image & Video Processing Multimedia Processing Art & Culture
28
mcp-flux-proOAuth
io.github.AceDataCloud
MCP server for Flux AI image generation
Image & Video Processing Multimedia Processing
23

Related MCP Servers

Wikimedia-Search-Images
Entertainment & Media
xiaobenyang-com
A
license
B
quality
D
maintenance
This MCP server enables AI assistants to search for images on Wikimedia Commons, providing detailed metadata and optional thumbnail combinations to assist AI models in visual comparisons.
Updated 8 months ago (2026-01-09 07:04 UTC)
1
2
Apache 2.0
brave-image-mcp
Search
ilker-tosun
F
license
A
quality
D
maintenance
An MCP server that provides image search capabilities via the Brave Image Search API, allowing AI assistants to search images with various filters and perform batch queries.
Updated 5 months ago (2026-04-08 17:26 UTC)
2
1
-
imagebank-mcp
Developer Tools Image & Video Processing Search
romulorasec
A
license
A
quality
C
maintenance
MCP server that allows AI coding agents to search, download, and convert stock photos from Pexels, Unsplash, and Pixabay into WebP format for direct use in web development projects.
Updated 2 months ago (2026-07-08 01:38 UTC)
3
25
2
MIT
MCP Image Resolver
Search Image & Video Processing
ahmaddioxide
A
license
A
quality
D
maintenance
An MCP server that enables AI assistants to search for royalty-free images from Pexels and Unsplash using natural language, returning structured results with metadata.
Updated 6 months ago (2026-03-05 09:44 UTC)
5
59
MIT

Latest Blog Posts

Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)
By Om-Shree-0709 on July 25, 2026.
mcp
Agent Identity
OAuth 2.1
Your AI Chatbot Just Exposed Your CEO's Salary to an Intern
By Om-Shree-0709 on July 2, 2026.
Agent Identity
MCP Security
OAuth Delegation
Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)
By Om-Shree-0709 on June 30, 2026.
Agentic Ai
Prompt Injection
WebAssembly

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Pexafy/pexafy-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server

pexafy-mcp

Nutzung (nichts zu installieren)

Claude (Web und Desktop)

Claude Code

Jeder andere MCP-Client

Was es kostet

Tools

search_photos – semantische Textsuche

search_photos_by_image – visuelle Suche anhand eines Beispielbilds

"das findest du herum" – mehr davon

Was kommt zurück

Selbst hosten

Konfiguration

Wie es funktioniert

Entwicklung

Lizenz

Available Tools

Output Schema

Output Schema

Output Schema

Tool Schema Changelog

Maintenance

Resources

Related MCP Connectors

Related MCP Servers

Latest Blog Posts

MCP directory API

`search_photos` – semantische Textsuche

`search_photos_by_image` – visuelle Suche anhand eines Beispielbilds