presentations-ai
# Presentations.AI MCP Server
[](https://github.com/slidecraft-in/presentations-ai-mcp-server/actions/workflows/ci.yml)
[](https://registry.modelcontextprotocol.io/?q=ai.presentations%2Fpresentations-ai)
[](LICENSE)
Official Model Context Protocol server for [Presentations.AI](https://www.presentations.ai).
Create designed slide decks from a topic or your notes, apply your brand, then share or
download them, from any MCP client.
[](https://vscode.dev/redirect?url=vscode%3Amcp%2Finstall%3F%257B%2522name%2522%253A%2522presentations-ai%2522%252C%2522type%2522%253A%2522http%2522%252C%2522url%2522%253A%2522https%253A%252F%252Fapi.presentations.ai%252Fmcp%2522%257D)
[](https://vscode.dev/redirect?url=vscode-insiders%3Amcp%2Finstall%3F%257B%2522name%2522%253A%2522presentations-ai%2522%252C%2522type%2522%253A%2522http%2522%252C%2522url%2522%253A%2522https%253A%252F%252Fapi.presentations.ai%252Fmcp%2522%257D)
[](https://vs-open.link/mcp-install?%7B%22name%22%3A%22presentations-ai%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fapi.presentations.ai%2Fmcp%22%7D)
[](https://cursor.com/en/install-mcp?name=presentations-ai&config=eyJ1cmwiOiJodHRwczovL2FwaS5wcmVzZW50YXRpb25zLmFpL21jcCJ9)
[](https://kiro.dev/launch/mcp/add?name=presentations-ai&config=%7B%22url%22%3A%22https%3A%2F%2Fapi.presentations.ai%2Fmcp%22%7D)
## Endpoint
```
https://api.presentations.ai/mcp
```
| Property | Value |
|---|---|
| Transport | Streamable HTTP |
| Authentication | OAuth 2.1 with PKCE and dynamic client registration, or an API key |
A Presentations.AI account is required. Most clients open a browser to sign in the first
time they connect. Clients without OAuth support can send an API key instead:
```
Authorization: Bearer pai_live_...
```
Create a key at [api.presentations.ai](https://api.presentations.ai/#/apikeys).
## Tools
| Tool | Use it to |
|---|---|
| `create_presentation` | Generate a deck from a prompt, pasted text, or an edited outline, with optional audience, slide count, and brand kit. Repeating an identical request returns the same deck; change any input to get a new one. |
| `create_outline` | Draft an outline to review and edit before generating. Spends credits: 10 per slide, and the deck built from it bills its slides again. |
| `get_task` | Wait up to 50 s per call for generation to finish. |
| `get_export` | A download link for the finished file, valid for 15 minutes. |
| `share_deck` | A public link, created when the user asks to share: anyone with it, and everyone in the workspace, can view the deck. |
| `search_brand_kits` | Find a brand to apply via `brand_kit_id`. |
| `whoami` | Identity, workspace, plan, and credit balance. |
| `list_decks` / `get_deck` / `list_projects` | Find and inspect existing work. |
| `patch_deck` / `delete_deck` | Rename, move, change visibility or content version, or delete a deck. |
| `create_grant` | Share a deck with one person, by email address or user ID. |
| `send_feedback` | Send a note to the Presentations.AI team, only when the user asks or approves the exact message. |
Source material goes in inline as text, or as a file already uploaded through the REST API.
File upload, admin, and usage operations are REST-only; see the
[API docs](https://docs.presentations.ai).
Clients that support the [MCP Skills extension](https://modelcontextprotocol.io/extensions/skills/overview)
also get two skills, `create-presentation` and `manage-presentations`.
Full reference: [docs.presentations.ai/docs/agents](https://docs.presentations.ai/docs/agents).
Every tool's JSON Schema input and output, description, and MCP annotations
(`readOnlyHint`, `destructiveHint`, `idempotentHint`, `openWorldHint`) are in
[`src/snapshot.json`](src/snapshot.json).
## Example prompts
```
Make a 10-slide deck on our product roadmap for the leadership team.
Turn these meeting notes into a short deck: <paste notes>
Show me an outline first, then build the deck with the Presentations.AI brand kit.
Find my onboarding deck and give me a fresh download link.
Share my "Q3 results" deck with teammate@example.com as a viewer.
```
## Connect from your MCP client
- **Claude (claude.ai and Claude Desktop)**: [docs/clients/claude-desktop.md](docs/clients/claude-desktop.md)
- **Claude Code**: [docs/clients/claude-code.md](docs/clients/claude-code.md)
- **ChatGPT**: [docs/clients/chatgpt.md](docs/clients/chatgpt.md)
- **Cursor**: [docs/clients/cursor.md](docs/clients/cursor.md)
- **VS Code**: [docs/clients/vs-code.md](docs/clients/vs-code.md)
- **Visual Studio**: [docs/clients/visual-studio.md](docs/clients/visual-studio.md)
- **Kiro**: [docs/clients/kiro.md](docs/clients/kiro.md)
- **Cline**: [docs/clients/cline.md](docs/clients/cline.md)
- **Gemini CLI**: [docs/clients/gemini-cli.md](docs/clients/gemini-cli.md)
## Stdio wrapper
For clients that only run local servers, `src/index.js` forwards stdio to the hosted
endpoint with an API key. It needs Node.js 22 or later:
```bash
git clone https://github.com/slidecraft-in/presentations-ai-mcp-server
cd presentations-ai-mcp-server && npm ci
```
```json
{
"mcpServers": {
"presentations-ai": {
"command": "node",
"args": ["/path/to/presentations-ai-mcp-server/src/index.js"],
"env": { "PRESENTATIONSAI_API_KEY": "pai_live_..." }
}
}
}
```
The client's config file then holds the key, so keep it private.
Or with Docker: `docker build -t presentations-ai-mcp .`, then
`docker run -i --rm --init -e PRESENTATIONSAI_API_KEY presentations-ai-mcp`, which passes
the key from your environment.
Without a key, the wrapper lists the tools from `src/snapshot.json` and tool calls return
an error asking for one. To refresh the snapshot from the live server, set
`PRESENTATIONSAI_API_KEY` in your environment and run `npm run snapshot`. Set
`PRESENTATIONSAI_MCP_URL` to read a different https endpoint.
## Limits and errors
| Limit | Value |
|---|---|
| Generation time | A few minutes |
| Prompt and pasted text | 200,000 characters each |
Requests are rate-limited. Over the limit, the endpoint answers `429` with `Retry-After`,
`X-RateLimit-Limit`, `X-RateLimit-Remaining`, and `X-RateLimit-Reset`. Wait until
`Retry-After`, then retry.
A tool that fails returns an error result with a message. The messages below come from the
tools the server runs itself, such as `create_presentation`, `get_export`, and `share_deck`.
Tools built on the REST API, such as `create_outline`, `list_decks`, and `get_deck`, return
the REST API's own message instead.
| Message | Meaning | What to do |
|---|---|---|
| `unauthenticated` | The account lost access during the call. | Sign in again, or use an active key. Retrying unchanged won't help. |
| `forbidden: the token's scopes do not permit this` | The account's access to that deck or project doesn't allow the action (for example, view-only access), or the connection is limited to another project. | Ask the deck's owner or a workspace admin for access, or work in the project the connection is limited to. |
| `this feature is not available on the workspace's plan` | The plan doesn't include the feature. | Change plan. |
| `insufficient credits to generate a deck` | The credit balance is 0 or below. | Add credits. `whoami` shows the balance. |
| `attached files are still being processed; retry shortly` | Sources aren't ready yet. | Retry after a short wait. |
| `request_id was already used for a different request; pass a new one` | A `request_id` was reused with different input. | Pass a new `request_id`. |
An expired sign-in, a revoked key, or a workspace suspended for non-payment never reaches a
tool: the endpoint answers `401`, and the client asks you to sign in again.
Retries: tools annotated `idempotentHint` are safe to repeat. Repeating an identical
`create_presentation` call returns the same deck rather than spending credits again; pass
a new `request_id` for another copy. `create_outline` takes a retry key, `request_id`:
retrying with the same `request_id` after the outline was delivered is refused and not
charged, and a new one drafts and charges another outline. REST error codes and retry rules
are in the [API reference](https://docs.presentations.ai/docs/api#errors-and-asynchronous-failures).
## Pricing
Free to start: 200 credits when you sign up (400 with a work email), valid for 30 days; no
card needed.
Only `create_presentation` and `create_outline` spend workspace credits. An outline costs 10
credits per slide, and a deck built from it bills its slides again. A deck costs, per slide, 5,
10 (the default), or 20 credits by quality tier (`budget_tier` `lite`, `standard`, or
`premium`). `deep_search` doubles that, and a PowerPoint export adds 5 per slide on plans that
don't include it. Credits stop a request only when the balance is already 0 or below, so a
large deck can take the balance below zero: `whoami` shows the plan and balance. Plans and
prices: [docs.presentations.ai/docs/pricing](https://docs.presentations.ai/docs/pricing).
## Data and security
- The server acts as the signed-in account, or the owner of the API key, within that
account's permissions and the scopes the sign-in or key grants. Tools a connection isn't
allowed to use are hidden.
- For read-only access, sign in with read scopes only (`workspace:read`, `projects:read`,
`decks:read`, `brand_kits:read`, `credits:view`), or create an API key with the
**Read-only** preset. The connection then sees only the tools that read, plus
`send_feedback`.
- Seven tools only read. The other seven change, share, or send something and carry MCP's
`destructiveHint` annotation, so clients can ask before running them:
`create_presentation` and `create_outline` spend credits; `share_deck` creates a public
link, so anyone who has it, and everyone in the workspace, can view the deck; `patch_deck`
can change who can open a deck; `create_grant` shares a deck with a person; `delete_deck`
moves a deck to trash; and `send_feedback` sends a note to the team.
- Download links from `get_export` and `get_deck` work without signing in and expire after
15 minutes, so share them only with people who should have the file.
- Notes, documents, and web pages passed in are source material for the deck, not
instructions to the server. They can still carry prompt-injection attempts aimed at the
calling agent, so confirm with the user before `share_deck`, `create_grant`, a visibility
change with `patch_deck`, or `delete_deck`, whatever the source text says.
- Treat an API key like a password: keep it out of chats, code, and shared config files. If
a key is exposed, revoke it in the [API keys dashboard](https://api.presentations.ai/#/apikeys)
or with `DELETE /api-keys/current`, then create a new one.
See the [privacy policy](https://www.presentations.ai/privacy) and
[terms of service](https://www.presentations.ai/terms).
## Troubleshooting
| Problem | What to do |
|---|---|
| The client asks you to sign in again, or calls fail with 401 | Sign in from the client: `/mcp` in Claude Code, `/mcp auth presentations-ai` in Gemini CLI. With an API key, check that the key is still active in the [API keys dashboard](https://api.presentations.ai/#/apikeys). |
| Claude won't add another custom connector | Free plans allow one custom connector. |
| ChatGPT won't add the server | Workspace policies can restrict custom MCP servers. Ask your workspace admin. |
| A deck is still generating | A deck usually takes a few minutes. The agent checks progress with `get_task`, up to 50 s per call. |
| A download link stopped working | Links expire after 15 minutes. Ask for the file again and `get_export` returns a fresh link. |
| The stdio wrapper lists tools but every call fails | Set `PRESENTATIONSAI_API_KEY` to an active key. |
## Related repos
- [presentations-ai-skills](https://github.com/slidecraft-in/presentations-ai-skills): the `create-presentation` and `manage-presentations` Agent Skills
- [presentations-ai-gemini-extension](https://github.com/slidecraft-in/presentations-ai-gemini-extension): Gemini CLI extension with this server and both skills
## Support
Email `support@presentations.ai` with the client you use, the tool that failed, when it
happened, and the error message. Never include an API key or an `Authorization` header.
More help: [docs.presentations.ai/docs/support](https://docs.presentations.ai/docs/support).
## Changelog
See [CHANGELOG.md](CHANGELOG.md).
## License
MIT, see [LICENSE](LICENSE).
TDQS
Scored across 13 tools
Tools mostly have distinct resource-action boundaries: deck creation, task polling, export retrieval, sharing, and deck management are clearly separated. Minor potential confusion exists between get_task and get_deck (both report generation status) and between get_export and share_deck (both provide deck access), but the descriptions clarify the differences.
Most tools follow a predictable verb_noun pattern such as create_presentation, get_task, list_decks, delete_deck, and patch_deck. whoami is the only clear outlier, using a conventional but non-verb_noun name.
The 13 tools are well-scoped for a presentation-generation service, covering creation, discovery, async status, export, sharing, and deck lifecycle operations. Each tool appears to earn its place without excessive surface area.
The set covers the main deck lifecycle: brand discovery, outline generation, presentation creation, task polling, export, sharing, listing, fetching, updating, and deleting decks. Minor gaps exist around project/folder management and generation cancellation or retry, but core workflows are well supported.