What this is#
MCP (the Model Context Protocol) is the open standard AI assistants use to call outside tools. Tagrly's MCP server exposes your analyzed photo library through it: one URL, one key, six tools. It is the same catalog the HTTP API serves, wearing the connector shape that AI clients set up in one step.
Every photo behind it has already been read by the analyzer: subject, scene, mood, a written description, finished alt text, and a hosted CDN URL. So when your assistant asks for "three candid shots of guests laughing at the reception", "a twilight exterior for the listing hero", or "friends at a table for the brunch post", what comes back is ready to embed, not a folder of filenames.
When the library does not have the shot, the curated tools say so with a gap signal instead of forcing a weak match. Your assistant can pass that straight to you as a shoot list.
Setup per client#
You need a workspace API key first: sign in and create one at Settings, API access (owners and admins only; the key is shown once). Then:
Claude Code#
claude mcp add --transport http tagrly https://tagrly.com/mcp \
--header "Authorization: Bearer tagrly_pk_YOUR_KEY"Claude (web and desktop)#
Settings, Connectors, Add custom connector. Enter
https://tagrly.com/mcp as the server URL and add a request
header named Authorization with the value
Bearer tagrly_pk_YOUR_KEY. Header-based connector auth is
rolling out on Claude's plans; if the field is not available on your
account yet, use Claude Code or the API connector below.
Cursor and other MCP clients#
{
"mcpServers": {
"tagrly": {
"url": "https://tagrly.com/mcp",
"headers": { "Authorization": "Bearer tagrly_pk_YOUR_KEY" }
}
}
}Claude API (MCP connector)#
"mcp_servers": [{
"type": "url",
"url": "https://tagrly.com/mcp",
"name": "tagrly",
"authorization_token": "tagrly_pk_YOUR_KEY"
}]The six tools#
| Tool | What it does | Cost |
|---|---|---|
search_photos | Instant keyword search. Filters for team, scene, mood, year, shot type and more. Returns compact records with the hosted URL and alt text. | Free |
find_photos | Plain-language search. A model interprets the sentence, resolves filters, and reranks the results. | 1 ask |
get_page_images | The page builder. Send a topic, get a curated set organized into slots with rewritten alt text and a gap signal. | 1 ask |
get_image | The full stored record for one photo, including the long description. | Free |
log_usage | Record which photos were rendered so future page calls can avoid repeats. | Free |
list_collections | The workspace's hand-curated collections. | Free |
The intended loop: get_page_images with the page topic,
render each pick's hosted_url with its
alt_rewrite, then log_usage with the ids you
embedded. On the next page, pass
exclude_recently_used_days and a 50-post series never
recycles a hero.
Authentication#
The server takes the same workspace API keys as the HTTP API, sent as
Authorization: Bearer tagrly_pk_.... Keys are created and
revoked at Settings, API access by a
workspace owner or admin, and each key reaches exactly one workspace's
photos. Discovery (connecting and listing tools) needs no key; every
tool call requires one.
What calls cost#
Four of the six tools are free and unmetered. The two model-curated
tools, find_photos and get_page_images, each
spend one ask from the workspace's monthly allowance, the same meter
the in-app AI search uses, and the same one the
pricing page describes. If a curated call
resolves without a model (for example a dry run), the ask is refunded.
MCP or the HTTP API?#
Same catalog, same auth, same behavior; the MCP server dispatches every tool call through the public API internally, so nothing can differ between the two. Choose by client: an AI assistant with connector support wants MCP (setup is one step and the tools are self-describing); a script, a WordPress build, or anything that speaks plain HTTP wants the API. The full text reference covers both.
Questions#
What is an MCP server for photos?
MCP is the open standard AI assistants use to call external tools. A photo MCP server exposes a photo library as callable tools: search by keyword, describe a shot in plain words, or request a curated set for a page. Tagrly's serves your own photos, already analyzed, with alt text written.
How do I connect my photo library to Claude?
Connect a Google Drive or Dropbox folder to Tagrly, create an API key
in Settings, and add https://tagrly.com/mcp as a connector
with the key in the Authorization header. Claude can then search and
place your photos itself.
Does it work with ChatGPT and Cursor?
Yes. It is standard MCP over Streamable HTTP with header auth; nothing about it is Claude-specific.
Can the server change my photos?
The tools read, curate and log usage. None of them delete photos, change your source folders, or touch your Drive or Dropbox.