Overleaf MCP Server
MCP server for interacting with Overleaf projects, enabling project listing, file reads, LaTeX section discovery, and revision-guarded writes through natural language.
README
olgitbridge + Overleaf MCP
This repository preserves the original Overleaf Git HTTP bridge and adds the MCP service used by the current deployment. The bridge talks to an Overleaf Community Edition or Pro instance through its web API, so it can run on a separate host.
The deployed code baseline is tagged deployed-20260801T090535Z. Site
configuration, credentials, bearer tokens, project IDs, synchronized papers,
runtime Git repositories, and auth-proxy state stay outside Git.
Capabilities
- Existing Git clone, pull, and push bridge on the configured Git HTTP port.
- Streamable HTTP MCP endpoint on loopback, with bearer authentication and Host/Origin checks.
- Server-side project allowlist addressed by stable
projectNamealiases. - Project and file listing, bounded text reads, LaTeX section discovery, and status summaries.
- Revision-guarded file and section writes with Overleaf read-back verification.
- Per-project serialization, timeout propagation, atomic tree replacement, interrupted-sync recovery, path confinement, and credential permission checks.
The MCP backend deliberately accepts only allowlisted project aliases. Raw Overleaf project IDs and Overleaf account credentials never come from MCP callers.
Repository layout
src/core/ Safe project, sync, credential, and Git services
src/transports/mcp-http.js Streamable HTTP MCP transport and tool schemas
src/olops/ Overleaf web operations
config/ Checked-in, non-secret configuration examples
test/ Unit and integration tests
docs/olgitbridge-mcp/ Design, implementation, and deployment notes
deploy/ Sanitized templates matching the deployed layout
Development
Production is currently exercised with Node.js 22 and npm 10.
npm ci
npm test
npm run check
The traditional Git bridge still reads config.js. Keep the tracked file
generic and apply the Overleaf URL and absolute persistent data directory only
in a release copy or deployment overlay.
To enable MCP, copy config/mcp-projects.example.json outside the repository,
create the referenced credential and token files with restrictive permissions,
and start the service with:
OLGITBRIDGE_MCP_CONFIG=/absolute/path/mcp-projects.json \
OLGITBRIDGE_CREDENTIAL_DIR=/absolute/path/credentials \
npm start
The MCP listener must remain on a loopback address. See
docs/olgitbridge-mcp/DEPLOYMENT.md and
deploy/README.md before exposing it through a separate
authentication proxy.
Selecting a project
Call list_projects first, then pass the returned key as projectName to
all project-specific tools:
{
"projectName": "example-paper",
"filePath": "main.tex"
}
Write tools also require the latest revision as expectedRevision. A stale
revision fails closed with REVISION_CONFLICT.
Git bridge compatibility
The Git HTTP endpoint remains compatible with the original form:
http://bridge.example.invalid/<overleaf-project-id>
The bridge performs whole-file synchronization through Overleaf's web API. Online editors may therefore report that a file changed externally after a Git or MCP write.
Upstream and license
This project derives from
axkibe/olgitbridge and retains its
AGPL-3.0-only license. The configured Git remote points at the maintainer's
fork; publishing remains an explicit operation.
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。