manim-server
An MCP server that enables creating and rendering Manim animations by building scenes with code sections and generating MP4 videos.
README
<div align="center">
manim-server
FastAPI + MCP server for trusted Manim script rendering. (uses Manim Community) <video src="https://github.com/user-attachments/assets/9be7e5de-c89a-4c5a-9b6e-2852ae104acf" controls="controls" muted="muted" class="d-block rounded-bottom-2 border-top width-fit" style="max-height:480px; min-height: 200px"> </video> </div>
Install
uv sync --all-groups
Configure
Create .env from the example:
cp .env.example .env
Supported keys:
HOST=127.0.0.1
PORT=8000
DATA_DIR=.manim-server-data
TEMPLATE_DIR=template
MANIM_CLI_FLAGS=-ql
MANIM_TIMEOUT_SECONDS=120
Notes:
DATA_DIRstores session JSON logs and rendered MP4 artifacts.TEMPLATE_DIRstores named full-wrapper Manim templates.MANIM_CLI_FLAGSis split like a shell command, then passed as subprocess args. Common values:-ql,-qm,-qh,-ql --fps 30.--save_sections,--media_dir, cache flags, scene path, and scene class are managed by the server.
Run
make run
Default server: http://127.0.0.1:8000.
Docker
Build the image:
make docker-build
Run it on http://127.0.0.1:8000:
make docker-run
Smoke-test the image:
make docker-smoke
The image is based on manimcommunity/manim:v0.20.1, matching the locked Manim dependency. That base image provides the Manim CLI plus common native rendering dependencies such as ffmpeg, Cairo/Pango, fonts, and minimal TeX Live. If a scene needs an extra TeX package, extend the image with that specific package instead of preinstalling all of TeX Live.
REST flow
The example builds one session from five Manim snippets, renders the final scene,
then downloads the MP4. Use Manim-Session-ID when you want a stable id; omit it
to let the server generate one.
# POST /sessions creates a new scene session.
# Response includes sessionId, title, sectionCount, and an empty section log.
curl -sS -X POST http://127.0.0.1:8000/sessions \
-H 'content-type: application/json' \
-H 'Manim-Session-ID: five-section-demo' \
-d '{"title":"Five animated sections curl smoke"}'
# POST /sessions/<sessionId>/section appends one Manim code section.
# render=false records code without invoking Manim; cache is ignored until rendering.
curl -sS -X POST http://127.0.0.1:8000/sessions/five-section-demo/section \
-H 'content-type: application/json' \
-d '{"title":"Create blue circle","code":"circle = Circle(color=BLUE)\nself.play(Create(circle), run_time=1.0)","render":false}'
# Another section. Sections run in append order inside the generated scene.
curl -sS -X POST http://127.0.0.1:8000/sessions/five-section-demo/section \
-H 'content-type: application/json' \
-d '{"title":"Create green square","code":"square = Square(color=GREEN).shift(RIGHT * 2)\nself.play(Create(square), run_time=1.0)","render":false}'
# Adds a third shape, still only updating the session JSON log.
curl -sS -X POST http://127.0.0.1:8000/sessions/five-section-demo/section \
-H 'content-type: application/json' \
-d '{"title":"Create yellow triangle","code":"triangle = Triangle(color=YELLOW).shift(LEFT * 2)\nself.play(Create(triangle), run_time=1.0)","render":false}'
# Adds a fourth animation section.
curl -sS -X POST http://127.0.0.1:8000/sessions/five-section-demo/section \
-H 'content-type: application/json' \
-d '{"title":"Create lower line","code":"line = Line(LEFT, RIGHT, color=RED).shift(DOWN * 1.5)\nself.play(Create(line), run_time=1.0)","render":false}'
# Final section render=true appends code and runs Manim. cache defaults to "use",
# so omitting it lets Manim reuse existing cached partial movie files.
curl -sS -X POST http://127.0.0.1:8000/sessions/five-section-demo/section \
-H 'content-type: application/json' \
-d '{"title":"Create purple ellipse","code":"ellipse = Ellipse(width=3, height=1.5, color=PURPLE).shift(UP * 1.5)\nself.play(Create(ellipse), run_time=1.0)\nself.wait(0.1)","render":true}'
# GET /sessions/<sessionId>/video downloads the full MP4 from the latest render.
curl -sS -o /tmp/five-sections-full.mp4 \
-w 'status=%{http_code} content_type=%{content_type} bytes=%{size_download}\n' \
http://127.0.0.1:8000/sessions/five-section-demo/video
Open returned latestRender.fullVideoUrl, or a section URL such as
/sessions/five-section-demo/sections/0001/video, in a browser or video client.
Cache modes for append_section with render=true, and for render_scene:
- Omit
cache, or set"cache":"use": default; let Manim reuse cache. "cache":"flush": delete Manim partial movie cache before rendering."cache":"disable": pass--disable_cachingto Manim for that render.
Session templates
POST /sessions accepts optional templateId:
curl -sS -X POST http://127.0.0.1:8000/sessions \
-H 'content-type: application/json' \
-d '{"title":"Lecture","templateId":"lecture"}'
Template assets are Python files under TEMPLATE_DIR/<templateId>.py.
By default this is ./template/<templateId>.py. Unknown missing templates
fall back to default.py; if default.py is absent, session creation fails
until the template directory is fixed.
Bundled templates:
default: plain title header.clean-title: title header with underline.dark-grid: dark background with faint coordinate grid.presentation-card: animated title card intro, then clears before sections.
Template files are full Manim scripts, not scene-body snippets. Keep the scene
class named GeneratedScene; Manim renders that class, and user sections are
appended at EOF at the construct() body indentation. Do not put dedented code
after construct() that would close that append point.
The reserved string literals "__SESSION_ID__", "__SESSION_TITLE__", and
"__TEMPLATE_ID__" are replaced before render. Include them where the template
needs those values:
from manim import *
from manim.opengl import *
class GeneratedScene(Scene):
def construct(self):
# DO NOT EDIT: replaced by manim-server before render.
session_id = "__SESSION_ID__"
session_title = "__SESSION_TITLE__"
template_id = "__TEMPLATE_ID__"
title = Text(session_title or "Untitled").to_edge(UP)
self.add(title)
# user sections append here
MCP endpoint
http://127.0.0.1:8000/mcp
Tools mirror REST: create_session, list_sessions, get_session, close_session, append_section, render_scene, reset_session.
Thanks
Special thanks to the Manim Community and 3Blue1Brown for their amazing work and inspiration.
Security
Scene code is trusted Python. Do not expose this server to untrusted users or networks.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。