Agents & MCP

The Prezent Platform API ships with a first-class Model Context Protocol (MCP) server, so AI agents can create and convert presentations as a tool call instead of writing custom HTTP code. Two ways to use it:

Hosted remote MCP at https://mcp.myprezent.com/mcp — OAuth 2.1, no install needed; works with Claude.ai Custom Connectors.
Local stdio server via pip install prezent-mcp-server — drop a config snippet into Claude Desktop, Cursor, Cline, Continue, or Zed and the tools appear in your agent's tool list.

The same eleven tools (presentation creation, status, download, template conversion, theme discovery, feedback) are exposed in both modes. The MCP server wraps the documented REST API — anything you can do through the API, an agent can do through MCP.

Source repository: github.com/prezentai/prezent-mcp.

Quick start — hosted MCP (Claude.ai)

The fastest path. Zero install.

In Claude.ai, open Settings → Custom Connectors → Add custom connector.
MCP server URL: https://mcp.myprezent.com/mcp
Click Connect. Claude.ai redirects through Prezent's OAuth login — sign in with your Prezent account.
The Prezent tools (create_presentation, convert_template, etc.) are now available in any Claude.ai chat.

That's it — your Prezent API key is fetched server-side from your Prezent account, so you don't paste it into Claude.

Quick start — local stdio (Claude Desktop / Cursor / Cline / Continue / Zed)

For locally-running agents and IDE integrations.

1. Install the server

pip install prezent-mcp-server

Requires Python 3.10+.

2. Get a Prezent API key

Contact your Customer Success Manager. The key is a Bearer token of the form pzbk_.... Set it as an environment variable.

3. Add the server to your agent's MCP config

Claude Desktop
Cursor
Cline / Continue
Zed

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "prezent": {
      "command": "prezent-mcp-server",
      "env": {
        "PRESENTATION_API_TOKEN": "pzbk_your_key_here"
      }
    }
  }
}

Restart Claude Desktop. The Prezent tools appear under the 🔌 plug icon in the chat input.

In Cursor, open Settings → Features → Model Context Protocol → Add new MCP server:

{
  "mcpServers": {
    "prezent": {
      "command": "prezent-mcp-server",
      "env": {
        "PRESENTATION_API_TOKEN": "pzbk_your_key_here"
      }
    }
  }
}

In your cline_mcp_settings.json (or Continue's ~/.continue/config.json mcpServers section):

{
  "mcpServers": {
    "prezent": {
      "command": "prezent-mcp-server",
      "env": {
        "PRESENTATION_API_TOKEN": "pzbk_your_key_here"
      }
    }
  }
}

In Zed's settings.json, under context_servers:

{
  "context_servers": {
    "prezent": {
      "command": {
        "path": "prezent-mcp-server",
        "args": [],
        "env": {
          "PRESENTATION_API_TOKEN": "pzbk_your_key_here"
        }
      },
      "settings": {}
    }
  }
}

4. Verify

In a fresh chat, ask the agent:

"Which Prezent templates can I use?"

The agent should invoke get_accessible_templates and return your company's themes.

Available tools

All tools target the Prezent Platform API. Async tools (presentation / template-conversion) return a callback_id that the agent polls via the matching check_*_status tool.

Tool	Purpose	Sync / Async
`create_presentation`	Start a presentation generation job. Accepts a prompt, optional template id, files / web links / historical context. Returns a `callback_id`.	Async
`check_presentation_status`	Poll the status of a `create_presentation` job. Returns slides, extracted images, theme/audience metadata when complete.	Sync
`download_presentation_tool`	Download the final `.pptx` (or `.pdf`) for a completed presentation.	Sync
`send_reaction_feedback`	Like/dislike a generated presentation.	Sync
`send_text_feedback`	Submit free-text feedback on a presentation.	Sync
`convert_template`	Start a Template Converter job that applies a target brand template to an uploaded deck. Returns a `callback_id`.	Async
`check_template_converter_status`	Poll a Template Converter job.	Sync
`send_template_converter_reaction_feedback`	Like/dislike a conversion.	Sync
`send_template_converter_text_feedback`	Submit free-text feedback on a conversion.	Sync
`get_accessible_templates`	List the brand templates / themes the caller's company has access to.	Sync
`validate_template`	Verify that a specific template id is accessible before kicking off a generation or conversion.	Sync

Each tool is wired to the equivalent REST endpoint documented in the interactive API Reference — same response envelope, same error catalog, same rate limits, same annual usage quotas.

Authentication

The hosted MCP server (mcp.myprezent.com) and the local stdio server authenticate the same calls differently:

Mode	Flow
Hosted	OAuth 2.1 authorisation code + PKCE. Claude.ai redirects to Prezent's Cognito-backed login, gets an access token (Cognito id_token), and presents it as `Bearer <token>` on every `/mcp` call. The remote-server middleware exchanges that for the user's Prezent API key behind the scenes — the API key never leaves Prezent infrastructure.
Local stdio	Bearer token from the `PRESENTATION_API_TOKEN` environment variable, sent directly on every API call.

Both modes ultimately hit the same Prezent Platform API endpoints described in this documentation. Auth, rate limits, error codes, and annual quotas behave identically.

Rate limits & quotas

Tool calls count against the same limits as direct API calls:

Gateway throttle — 10 req/s, 5 burst, 1,000 req/day per key.
Annual quotas — slide-generation (create_presentation) and presentation-download (download_presentation_tool) consume the per-key annual quotas (default 50,000 / 1,000,000).
Per-minute throttle — applied per company × category when configured.

See the Developer Guide → Usage quotas and rate limits for the full picture.

Discovery + interop

The hosted server implements the relevant MCP / OAuth interop standards:

MCP transport: Streamable HTTP at https://mcp.myprezent.com/mcp (current MCP-spec default; Server-Sent Events for streaming responses).
OAuth 2.1 authorisation server metadata: GET https://mcp.myprezent.com/.well-known/oauth-authorization-server (RFC 8414).
Protected-resource metadata: GET https://mcp.myprezent.com/.well-known/oauth-protected-resource (RFC 9728).
Dynamic client registration: POST https://mcp.myprezent.com/oauth/register (RFC 7591).
Liveness: GET https://mcp.myprezent.com/healthz.

Most agent clients (Claude.ai, Cursor, Cline) handle this discovery automatically when you give them the MCP server URL.

Source, issues, and roadmap

GitHub: github.com/prezentai/prezent-mcp
Issues / feature requests: file them on the GitHub repo.
PyPI: prezent-mcp-server
Versioning: the MCP server tracks the OpenAPI contract version (currently 1.0.0). New tools are additive; tool signatures don't change without a major-version bump.

Where to go next

The interactive API Reference — every underlying endpoint with try-it-out forms.
Getting Started — make your first raw HTTP call (useful for debugging what an MCP tool is doing).
AutoGenerator — full request/response detail for the presentation-creation flow that create_presentation wraps.
Template Converter — full request/response detail for convert_template.

Quick start — hosted MCP (Claude.ai)​

Quick start — local stdio (Claude Desktop / Cursor / Cline / Continue / Zed)​

1. Install the server​

2. Get a Prezent API key​

3. Add the server to your agent's MCP config​

4. Verify​

Available tools​

Authentication​

Rate limits & quotas​

Discovery + interop​

Source, issues, and roadmap​

Where to go next​