Pinterest Opens Taste Graph to Third-Party Agents via MCP

On June 17, 2026, Pinterest announced new agentic advertising tools anchored by the launch of the Pinterest Model Context Protocol (MCP). The release allows developers and marketers to route Pinterest’s proprietary consumer intent signals directly into external third-party AI workflows, rather than requiring users to operate exclusively within Pinterest’s native dashboard.

Model Context Protocol Infrastructure

The Pinterest MCP operates as a standardized server layer, based on the open-source protocol originally introduced by Anthropic. It provides a secure bridge for AI agents and enterprise copilots to query live campaign performance data, account analytics, and keyword insights.

Crucially, the server exposes data from the Pinterest Taste Graph. This dataset is trained on billions of visual and text signals to map emerging consumer trends. By exposing this via MCP, advertisers can prompt models like Claude or ChatGPT to analyze campaign gaps against live platform trends.

The system is currently deployed in a read-only state. External tools can extract and analyze data, but they cannot execute configuration changes or launch campaigns without human intervention. Initial alpha access is restricted to select agency and ad-tech partners, including PMG, Pacvue, Dentsu, Havas, Innovid by Mediaocean, and Omnicom’s Jump450.

Performance+ Creative Selection

Alongside the infrastructure update, Pinterest released a new AI model for its Performance+ suite, a product line that now accounts for roughly 30% of the platform’s lower-funnel revenue.

Previous optimization models functioned at the campaign or ad-group level. The new Performance+ creative model evaluates a complete library of merchant assets and dynamically selects the specific variant optimized for each individual ad impression. Internal benchmarks from Pinterest indicate this impression-level variant selection yields a 7.5% increase in click volume compared to standard single-variant delivery.

Pinterest also launched a closed beta of Business Assistant for select U.S. advertisers. This in-platform tool sits within Pinterest Ads Manager, visualizing breakout category trends and generating specific promotion recommendations based on historical account data.

Agentic Discovery Sandbox

For consumers, Pinterest introduced Ask Pinterest, a standalone experimental mobile application currently restricted to the United States.

The app replaces standard keyword search with a persistent conversational interface. Users engage in multi-step dialogues for complex planning tasks, such as interior design or event preparation. The underlying model references the user’s historical saved Pins and Boards as context for its recommendations, treating search as a continuous session rather than isolated queries.

For developers building ad-tech integrations, the Pinterest MCP establishes a clear architectural pattern. If your application relies on aggregating cross-platform media insights, standardizing on the Model Context Protocol now allows you to ingest Pinterest data without building bespoke API polling mechanisms.