Intelligence API — Thisispaper

&larr;Intelligence<br>Thisispaper is an independent art, architecture and design publication.<br>Every project in the archive was curated by human editors since 2011.

Try it in 30 seconds

The MCP endpoint answers anonymously, no key and no signup. Anonymous calls are<br>capped at 5 results; a free key (200 calls/month) unlocks full results.

curl -X POST https://www.thisispaper.com/intelligence/api/mcp \<br>-H "Content-Type: application/json" \<br>-d '{"jsonrpc":"2.0","id":1,"method":"tools/call",<br>"params":{"name":"search_projects",<br>"arguments":{"query":"concrete chapel"}}}'<br>Or skip curl entirely: add the server to your AI client<br>and ask it to search Thisispaper.

For AI agents (MCP)

The catalog is a remote Model Context Protocol server<br>(Streamable HTTP, stateless), so Claude, ChatGPT and other AI agents can query it natively:

https://www.thisispaper.com/intelligence/api/mcp<br>Without a key the tools run in anonymous preview mode, capped at 5 results.<br>Add Authorization: Bearer tip_live_... to the connector for full,<br>metered access. An agent can even mint its own key mid-conversation with<br>get_api_key.

The server exposes nine tools:<br>TOOL search_projects<br>Search the curated catalog by keywords. Args: query (required), limit (max 20). Returns ranked project summaries: name, studio, city, year, category, intro, thumbnail, URL, citation_url, provenance.

TOOL search_by_taste

Search by taste metadata: material, light, color,<br>mood, composition, subject.<br>Unique to Thisispaper: these tags were written during human curation. Pass at least<br>one dimension, e.g. {"material": "rammed earth", "light": "dusk"}.<br>Results are ranked by how many dimensions match, then by recency.

TOOL browse<br>Newest projects first. Optional category filter: architecture, interiors, design, photography or art.

TOOL discover<br>One project chosen at random from the archive. Editorial serendipity, the way a reader opens a magazine to an unplanned spread. No arguments.

TOOL similar_projects<br>Visual + semantic neighbours for a project slug (precomputed SigLIP neighbours).

TOOL get_project<br>Full metadata for one slug: description, design language, details, taste dimensions, URL, citation URL, provenance.

TOOL get_api_key<br>Mints a free API key (200 calls/month) for your email, right from the conversation. The key is shown once. Max 3 keys per email.

TOOL search

Connector-contract search for clients that follow the ChatGPT connector spec.<br>Takes a query, returns results as id, title and url. ChatGPT calls this tool in<br>normal chat.

TOOL fetch

Connector-contract fetch: takes an id from search and returns the full record with<br>title, text, url and metadata. The result also embeds the project's real thumbnail<br>image (up to 2 MB), which image-capable clients render inline.

Any MCP client works; no session or OAuth handshake is required.

Add to your client<br>claude.ai / Claude Desktop

Settings &rarr; Connectors &rarr; Add custom connector, then paste the server URL.<br>Connectors added on claude.ai sync to Claude Desktop.

Claude Code<br>claude mcp add --transport http thisispaper \<br>https://www.thisispaper.com/intelligence/api/mcp<br>ChatGPT

Settings &rarr; Connectors &rarr; add the server URL. Works in normal chat through<br>the search and fetch tools.

Cursor<br>One-click install,<br>or add to .cursor/mcp.json:

"mcpServers": {<br>"thisispaper": {<br>"url": "https://www.thisispaper.com/intelligence/api/mcp"<br>Codex CLI<br>In ~/.codex/config.toml:<br>[mcp_servers.thisispaper]<br>url = "https://www.thisispaper.com/intelligence/api/mcp"<br>Gemini CLI<br>gemini extensions install https://github.com/zaxarovcom/thisispaper-mcp<br>npm<br>A local stdio bridge to the same server, for clients without HTTP transport:<br>npx thisispaper-mcp

Any other client that speaks Streamable HTTP: point it at the URL above, or bridge<br>via npx mcp-remote https://www.thisispaper.com/intelligence/api/mcp.

Tested with Claude, ChatGPT, Codex, Grok, Gemini and Cursor.

Prompts worth stealing<br>Once the connector is added, ask your agent in plain language. Lines that work:<br>Architecture"houses that stay cool without machines"<br>Architecture"concrete chapels in the Alps"<br>Photography"documentary photographers of vanishing ways of life"<br>Photography"portraits of remote communities"<br>Art"installations that use light as material"<br>Art"sculpture in aluminium"<br>Design"chairs that made minimalism warm"<br>Design"furniture in raw aluminium"<br>Interiors"apartments under 65 square meters that feel generous"<br>Taste search"material rammed earth, light dusk"<br>Taste search"mood melancholic, composition frontal"

REST API<br>The same catalog over plain GET endpoints. Base URL:<br>https://www.thisispaper.com/intelligence/api/v1<br>All endpoints return JSON and are CORS-open for browser apps. REST calls require<br>an API key (free, self-serve below); pass it as a bearer<br>token, or ?key= for quick tests:

curl "https://www.thisispaper.com/intelligence/api/v1/search?q=concrete+chapel" \<br>-H "Authorization: Bearer tip_live_..." Every response carries...