crepe-mcp
An MCP server that turns an AI agent into a slide author, enabling drafting decks in Pandoc Markdown, compiling to PDF or PowerPoint, validating with PNG exports, and pulling in research from the web, Wikipedia, and Semantic Scholar.
README
CREPE — Compile, Research, Export, Presentation Engine
An MCP server for Goose that turns an AI agent into a capable slide author: draft decks in Pandoc Markdown, compile to Beamer PDF or PowerPoint, validate visually with PNG exports, and pull in research from the web, Wikipedia and Semantic Scholar.
Tools (12 total)
Presentation (stateful)
| Tool | Purpose |
|---|---|
create_presentation |
Start a new deck, set all metadata |
get_presentation |
Inspect metadata + slide list |
get_slide |
Read a single slide's full content |
set_slide |
Add or replace a slide (index < len → replace, else append) |
compile_presentation |
Compile to PDF (Beamer/lualatex) or PPTX (pandoc) |
render_slides_as_pngs |
Render compiled artifact to PNG sequence for validation |
cleanup_presentation |
Delete a presentation's in-memory state and on-disk scratch dir |
Research & utilities
| Tool | Purpose |
|---|---|
academic_search |
Semantic Scholar (no key needed) |
web_search |
Tavily API web search |
wikipedia_search |
Wikipedia article search |
wikipedia_read |
Full Wikipedia article text |
fetch_webpage |
Extract text from any URL (headless browser or urllib) |
Installation
# From PyPI (once published)
uv tool install crepe-mcp
# From source
git clone https://github.com/mariolpantunes/crepe-mcp
cd crepe-mcp
uv tool install .
System requirements
- pandoc — slide compilation
- lualatex — PDF/Beamer output (
texlive-fullor MacTeX) - pymupdf — PDF→PNG rendering (installed via pip, no system dep)
- PPTX→PNG rendering — platform-dependent:
- Linux: LibreOffice is required (native
soffice/libreofficeon PATH, or installed as a Flatpak —flatpak install flathub org.libreoffice.LibreOffice, auto-detected at runtime, no config needed). There is no fallback on Linux: aspose-slides' embedded .NET runtime dynamically loadslibssl.so.1.1, which distros that ship only OpenSSL 3 (Slackware, recent Arch, etc.) no longer have, and no newer aspose-slides release has moved off it. Without LibreOffice,render_slides_as_pngs(format="pptx") returns a clean error instead of a crash. - macOS (and other platforms): LibreOffice is preferred if found (native
binary or
/Applications/LibreOffice.app); otherwise falls back to aspose-slides (installed via pip — no OpenSSL 1.1 issue on macOS). Aspose is a commercial library; withoutCREPE_ASPOSE_LICENSE_PATHit runs in evaluation mode and watermarks output PNGs.
- Linux: LibreOffice is required (native
Environment variables
All variables are prefixed with CREPE_ to avoid collisions.
| Variable | Required | Purpose |
|---|---|---|
CREPE_TAVILY_API_KEY |
No | Enables web_search via Tavily; graceful warning if absent |
CREPE_HEADLESS_BROWSER_PATH |
No | Path to Chromium-compatible browser for fetch_webpage. See macOS Browser Paths below. Falls back to urllib if unset. |
CREPE_LIBREOFFICE_PATH |
No | Path to a soffice/libreoffice executable, overriding auto-detection. Required on Linux only if auto-detection (PATH, Flatpak) fails. |
CREPE_ASPOSE_LICENSE_PATH |
No | Path to Aspose .lic file for watermark-free PPTX→PNG export on the aspose fallback path (macOS/other, when LibreOffice isn't found). Evaluation mode (with watermarks) is used if unset. |
macOS & Headless Browser Setup
To enable JavaScript-rendered webpage fetching via fetch_webpage on macOS, set CREPE_HEADLESS_BROWSER_PATH to any Chromium-based browser application path:
- Google Chrome:
/Applications/Google Chrome.app/Contents/MacOS/Google Chrome - Brave Browser:
/Applications/Brave Browser.app/Contents/MacOS/Brave Browser - Microsoft Edge:
/Applications/Microsoft Edge.app/Contents/MacOS/Microsoft Edge
Note on Safari: Apple's native Safari (
safaridriver) does not support CLI--headlessmode for DOM dumping (--dump-dom). If you only have Safari installed or choose not to setCREPE_HEADLESS_BROWSER_PATH,fetch_webpageautomatically falls back to pure Pythonurllib+ HTML stripping, which works instantaneously for all static articles, documentation, and Wikipedia pages.
Goose integration
You can automate this entire configuration using our built-in interactive setup script:
# Auto-detect browsers/keys, export profile variables, and register in ~/.config/goose/config.yaml
./setup.py --install
# To remove and clean up later:
./setup.py --uninstall
Alternatively, you can add crepe-mcp manually to ~/.config/goose/config.yaml using one of the options below.
Use envs (literal values), not env_keys. Goose resolves every env_keys
entry against its own keyring/secrets store — it does not fall back to a
plain shell environment variable of the same name. All of CREPE's env vars are
optional, and declaring one under env_keys without a matching Goose secret
makes the whole extension fail to start with Failed to fetch secret ... not found. Only fill in the values you actually need (all are optional; omit any
you don't use):
Option 1: Run directly from your local clone (Recommended for testing)
If you cloned the repository locally (e.g. to ~/git/crepe-mcp), you can tell Goose to execute it directly via uv run without installing it globally:
extensions:
crepe:
name: crepe
type: stdio
cmd: uv
args: ["--directory", "/home/username/git/crepe-mcp", "run", "crepe-mcp"]
envs:
CREPE_TAVILY_API_KEY: ""
CREPE_HEADLESS_BROWSER_PATH: ""
CREPE_ASPOSE_LICENSE_PATH: ""
CREPE_LIBREOFFICE_PATH: ""
(Replace /home/username/git/crepe-mcp with your actual absolute path to the repository).
Option 2: Run as an installed tool (uv tool install .)
If you ran uv tool install . inside the project directory (or installed via pip), crepe-mcp is on your PATH:
extensions:
crepe:
name: crepe
type: stdio
cmd: crepe-mcp
args: []
envs:
CREPE_TAVILY_API_KEY: ""
CREPE_HEADLESS_BROWSER_PATH: ""
CREPE_ASPOSE_LICENSE_PATH: ""
CREPE_LIBREOFFICE_PATH: ""
Option 3: Run ephemerally via uvx (Once published to PyPI)
If running directly from PyPI without local installation:
extensions:
crepe:
name: crepe
type: stdio
cmd: uvx
args: ["crepe-mcp"]
envs:
CREPE_TAVILY_API_KEY: ""
CREPE_HEADLESS_BROWSER_PATH: ""
CREPE_ASPOSE_LICENSE_PATH: ""
CREPE_LIBREOFFICE_PATH: ""
Option 4: Add via Goose CLI (goose configure)
If you prefer Goose's interactive terminal setup instead of editing config.yaml:
- Run
goose configurein your terminal. - Select
Add Extension→Command-line Extension (stdio). - For Name, enter:
crepe - For Command, enter either:
crepe-mcp(if installed globally)- OR
uv --directory /path/to/crepe-mcp run crepe-mcp(for local repo)
- When prompted for environment variables, add them as literal values
(
CREPE_TAVILY_API_KEY, etc.) — all are optional, so only add the ones you use. Ifgoose configureonly offers a secrets/keyring option rather than a plain value, prefer editingconfig.yamldirectly (Options 1-3 above) to useenvsinstead.
Typical agent workflow
# Build a deck
create_presentation(title="My Talk", author="Ada Lovelace")
set_slide(id, 0, "Introduction", "- Point one\n- Point two")
set_slide(id, 1, "Results", "$$E=mc^2$$")
# Validate PDF layout
compile_presentation(id, "/tmp/deck.pdf", format="pdf")
render_slides_as_pngs(id, format="pdf") # → slide_001.png …
# Deliver as PPTX
compile_presentation(id, "/tmp/deck.pptx", format="pptx")
render_slides_as_pngs(id, format="pptx") # validate PPTX layout
Pandoc slide syntax quick reference
| Element | Syntax |
|---|---|
| Section divider | # Section Title as slide content |
| Bullet | - item |
| Incremental bullet | > - item |
| Math block | $$ E=mc^2 $$ |
| Code | ```python\ncode\n``` |
| Image |  |
| Speaker notes | ::: notes\ntext\n::: |
| Two-column layout | :::: {.columns}\n::: {.column width="50%"}\nLeft\n:::\n::: {.column width="50%"}\nRight\n:::\n:::: |
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
E2B
Using MCP to run code via e2b.