presskit-mcp
An MCP server for publishing markdown articles to Medium and Substack via unofficial APIs, with tools to list, create, and manage posts and drafts.
README
presskit-mcp
A Python FastMCP server and CLI for publishing to Medium and Substack.
Stability warning: Both integrations rely on unofficial/deprecated access paths. Medium's REST API is frozen (no new tokens issued). Both platforms' internal endpoints are undocumented and may change without notice.
Install
pip install -e .
This installs two commands:
presskit— CLI for direct publishing from the terminalpublishing-mcp— MCP server for use with Claude Desktop/Claude Code
CLI Usage (presskit)
Publish a markdown file
# Publish to Medium as a draft
presskit publish medium --file docs/drafts/my-article.md
# Publish to Substack as a draft
presskit publish substack --file docs/drafts/my-article.md
# Publish to both platforms at once
presskit publish both --file docs/drafts/my-article.md
# Publish live (not draft)
presskit publish medium --file article.md --status public --tags "python,automation"
# Substack: publish to web only (no subscriber email)
presskit publish substack --file article.md --status public --no-email
The CLI reads YAML frontmatter from the markdown file:
---
title: "My Article Title"
subtitle: "Optional subtitle for Substack"
tags: [python, automation, infrastructure]
---
# My Article Title
Article body here...
If no frontmatter title is present, the first # Heading is used.
List posts
# List your Medium posts
presskit list medium --username alexander.g.moore1
# List Substack posts
presskit list substack --subdomain alexgmoore
# List Substack drafts
presskit drafts substack
CLI options reference
presskit publish <platform> [options]
--file, -f Markdown file to publish (required)
--status, -s draft | public | unlisted (default: draft)
--tags, -t Comma-separated tags, e.g. "python,devops" (Medium)
--subtitle Post subtitle (Substack)
--no-email Publish to web only, skip subscriber email (Substack)
presskit list <platform> [options]
--username, -u Medium username without @ (Medium)
--subdomain, -d Substack subdomain (Substack)
--limit, -n Max results (default: 10)
presskit drafts substack
MCP Server Usage
Run the server
# Direct
python server.py
# Or via installed entry point
publishing-mcp
Add to Claude Code (project-level)
Create .claude/mcp.json in your project:
{
"mcpServers": {
"publishing": {
"command": "python3",
"args": ["/absolute/path/to/presskit-mcp/server.py"],
"env": {
"MEDIUM_SESSION_COOKIE": "your_sid_cookie",
"MEDIUM_AUTH_STATE_FILE": "/path/to/medium-auth.json",
"SUBSTACK_EMAIL": "you@example.com",
"SUBSTACK_PASSWORD": "your_substack_password",
"SUBSTACK_PUBLICATION_URL": "https://yourpub.substack.com"
}
}
}
}
Add to Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (Mac):
{
"mcpServers": {
"publishing": {
"command": "python",
"args": ["/absolute/path/to/presskit-mcp/server.py"],
"env": {
"MEDIUM_SESSION_COOKIE": "your_sid",
"MEDIUM_AUTH_STATE_FILE": "/path/to/medium-auth.json",
"SUBSTACK_EMAIL": "you@example.com",
"SUBSTACK_PASSWORD": "your_password",
"SUBSTACK_PUBLICATION_URL": "https://yourpub.substack.com"
}
}
}
}
Restart Claude Desktop/Code after adding the config.
FastMCP Development & Testing
presskit-mcp is built on FastMCP. Here's how to work with it during development.
Inspect tools interactively
The MCP Inspector opens a browser UI where you can call any tool, see inputs/outputs, and debug:
# Using the mcp CLI (installed with mcp[cli])
mcp dev server.py
This starts the server and opens an interactive inspector at http://localhost:5173. You can:
- Browse all 15 registered tools
- Fill in parameters and execute them
- See JSON responses in real-time
- Test error handling
Run the server with stdio transport (default)
# FastMCP defaults to stdio transport (what Claude Desktop/Code expects)
python server.py
Run with SSE transport (for remote/HTTP access)
# Start as an HTTP server on port 8000
mcp run server.py --transport sse --port 8000
Then connect from any MCP client using http://localhost:8000/sse.
Call a tool directly via mcp call
# One-shot tool invocation without starting a persistent server
echo '{"username": "alexander.g.moore1", "limit": 5}' | \
mcp call server.py medium_list_posts
List all registered tools
mcp tools server.py
Environment variables for testing
# Medium (session-based — no integration token needed)
export MEDIUM_SESSION_COOKIE="your_sid_value"
export MEDIUM_AUTH_STATE_FILE="/path/to/medium-auth.json"
# Medium (REST API — only if you have an existing token)
export MEDIUM_INTEGRATION_TOKEN="your_token"
# Substack
export SUBSTACK_EMAIL="you@example.com"
export SUBSTACK_PASSWORD="your_password"
export SUBSTACK_PUBLICATION_URL="https://yourpub.substack.com"
Running tests
python3 -m unittest discover -s tests -v
Tests use sys.modules patching to stub third-party deps — no external services needed.
Tools Reference
Medium (6 tools)
| Tool | Auth | Method |
|---|---|---|
medium_get_current_user |
Integration token | REST API |
medium_get_publications |
Integration token | REST API |
medium_create_post |
Integration token | REST API |
medium_create_post_session |
Session cookie | GraphQL + Delta OT |
medium_list_posts |
Session cookie | Unofficial GraphQL |
medium_get_post_stats |
Session cookie | Unofficial GraphQL |
Substack (9 tools)
| Tool | Auth | Method |
|---|---|---|
substack_get_publication_info |
Email/password or cookie | python-substack |
substack_get_all_publications |
Email/password or cookie | python-substack |
substack_list_posts |
None (public) | Raw HTTP |
substack_get_post |
None (public) | Raw HTTP |
substack_search_publications |
None (public) | Raw HTTP |
substack_get_subscriber_count |
Email/password or cookie | python-substack |
substack_list_drafts |
Email/password or cookie | python-substack |
substack_create_draft |
Email/password or cookie | python-substack |
substack_publish_post |
Email/password or cookie | python-substack |
Credentials
Medium session cookie
Medium no longer issues API tokens. The session-based tools use browser cookies:
- Log in to medium.com in a browser
- Extract the
sidcookie from DevTools → Application → Cookies - For full functionality, save the complete browser state with
playwright-cli state-save medium-auth.jsonand setMEDIUM_AUTH_STATE_FILE
Substack
Use your Substack email and password directly. If your account uses magic links only:
- Sign out of Substack
- Click "Sign in with password"
- Click "Set a new password"
Architecture
presskit-mcp/
├── server.py # FastMCP server entry point (15 tools)
├── cli.py # CLI entry point (presskit command)
├── medium/
│ ├── client.py # REST API + GraphQL + Delta OT HTTP layer
│ └── tools.py # 6 MCP tools with Pydantic input models
├── substack/
│ ├── client.py # python-substack wrapper + raw HTTP
│ └── tools.py # 9 MCP tools with Pydantic input models
├── tests/
│ └── test_static.py # Unit tests (no external deps needed)
├── pyproject.toml # pip installable, entry points
└── config.yaml # Credential reference (not loaded by code)
How Medium session publishing works
Medium's web editor saves content via a delta-based Operational Transform system — not GraphQL or REST. The medium_create_post_session tool replicates this:
- GraphQL
createPost→ creates an empty draft, returnspost_id POST /p/{post_id}/deltas→ writes title + paragraphs as OT deltas- GraphQL
setPostTags→ sets up to 5 tags - GraphQL
publishPost→ publishes (optional)
This was discovered by decompiling Medium's Android APK and capturing network traffic from the web editor.
Known limitations
- Medium session cookies expire — you'll need to refresh
medium-auth.jsonperiodically (re-login via browser) - Medium REST API — No new integration tokens. Only existing holders can use
medium_create_post - Medium inline formatting — Bold/italic markup positions in the delta API need accurate character offsets; the current markdown parser sends plain text paragraphs
- Substack — No official API. All endpoints are reverse-engineered. Keep requests under 1/sec
- Images — Neither platform supports programmatic inline image upload via these tools yet
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。