substack-mcp

FastMCP server for Substack: publish Notes and posts, pull analytics, manage drafts, bridge Obsidian vault drafts to Substack, and generate visual cards with pluggable image generators.

Quick reference

substack__test_connection             -- verify auth (start here)
substack__publish_note text=...       -- publish a Note immediately
substack__list_vault_drafts           -- show drafts from your vault file
substack__publish_vault_draft index=N -- publish one vault draft as a Note
substack__create_draft title=... body=... -- create a post draft
substack__get_dashboard days=30       -- subscriber + view KPIs
substack__capture_analytics_to_vault  -- snapshot analytics to vault markdown

Tool inventory (30)

Notes + vault pipeline

• test_connection(publication?) -- verify auth, return your profile
• list_publications() -- show configured publications and which is default
• publish_note(text, publication?, attachment_ids?) -- publish a Note immediately; markdown supported
• create_note_attachment(image_path?, image_url?, link_url?, publication?) -- upload image or register link; returns attachment UUID for use with publish_note
• list_my_notes(limit?, publication?) -- your recent Notes with reaction + comment counts
• reply_to_note(note_id, text, publication?) -- reply to a Note by ID
• list_vault_drafts() -- parse vault drafts file, return index + preview for each draft
• publish_vault_draft(index, publication?, move_to_published?) -- publish one vault draft as a Note
• batch_publish_vault_drafts(indices, publication?) -- publish multiple vault drafts in sequence

Post management

• create_draft(title, body, subtitle?, audience?, publication?, section_id?) -- create a post draft; body is markdown
• update_draft(draft_id, title?, body?, subtitle?, audience?, publication?, section_id?) -- edit an existing draft
• publish_post(draft_id, send_email?, audience?, publication?, section_id?) -- publish a draft live to subscribers
• schedule_post(draft_id, publish_at, publication?, section_id?) -- schedule a draft for future publication (ISO 8601)
• list_sections(publication?) -- list configured sections for a publication; required to publish on publications with sections
• list_drafts(limit?, publication?) -- list unpublished drafts
• list_published(limit?, publication?) -- list published posts with basic stats
• get_post(identifier, publication?) -- get full post by slug or numeric ID
• upload_image(image_path, publication?) -- upload an image to Substack CDN; returns CDN URL
• react(post_id, publication?) -- heart a post
• restack(post_id, publication?) -- restack a post
• comment(post_id, body, publication?) -- comment on a post
• get_feed(limit?, publication?) -- your reader feed (posts from publications you follow)

Analytics

• get_dashboard(days?, publication?) -- KPIs: total/paid subscribers, views, growth, ARR
• get_post_stats(post_id, publication?) -- views, opens, clicks, shares, conversions for one post
• get_subscriber_growth(publication?) -- subscriber count over time
• get_growth_sources(publication?) -- subscriber growth by source (search, recommendations, direct, social)
• get_top_posts(limit?, publication?) -- posts ranked by engagement
• get_earnings(publication?) -- revenue data for paid publications
• get_recommendation_stats(direction?, publication?) -- recommendation network performance
• capture_analytics_to_vault(publication?) -- write a weekly analytics snapshot to your vault

Install

Open Claude Code, paste:

/plugin marketplace add adelaidasofia/substack-mcp
/plugin install substack-mcp@substack-mcp

git clone https://github.com/adelaidasofia/substack-mcp
cd substack-mcp
pip3 install -r requirements.txt
python3 -c "import server; print('OK')"

Configuration

Copy config.example.json to config.json and fill in your values:

cp config.example.json config.json

config.json is gitignored. Key fields:

Session cookie setup

Substack does not have a public API with OAuth. Auth uses your browser session cookie.

1. Open Chrome and log in to Substack.
2. Open DevTools (F12) and go to the Application tab.
3. Under Cookies, find substack.com.
4. Copy the value of substack.sid.
5. Paste it into config.json under the matching publication's cookie field.

The raw value works; the server accepts both abc123 and substack.sid=abc123.

Session cookies expire. If tools return {"error": "Auth failed"}, re-extract the cookie.

Register in Claude Code

Add to your .mcp.json (project-scoped) or via claude mcp add -s user:

{
  "mcpServers": {
    "substack": {
      "command": "python3",
      "args": ["/path/to/substack-mcp/server.py"]
    }
  }
}

Restart Claude Code after editing .mcp.json. Verify with claude mcp list.

Vault integration

Set vault_drafts_path in config.json to a markdown file in your Obsidian vault. Format each draft as a section separated by ---. Sections under ## Ready to Post are surfaced first by list_vault_drafts. Sections under ## Essay Seeds come next.

After publishing, publish_vault_draft moves the draft to a ## Published section and appends the Substack URL and timestamp.

Pluggable image generators

Visual card generation is handled by the adapter set in config.json under image_generator.default.

Shipped adapters

pillow_local (default) -- pure Pillow, no external API. Renders 1080x1080 PNG cards locally using fonts from the fonts/ directory. Three pillar templates: warm mustard with optional figure (P1), deep burgundy bold quote (P2), mustard with section tag (P3). Use with render_card.py as a standalone CLI.

canva -- Canva MCP choreography. Returns a steps list describing the Canva MCP tool calls needed to clone a template, replace text, replace the figure, and export PNG. The Claude session executing the playbook follows these steps. Requires Canva MCP connected in Claude Code and design IDs filled in under image_generator.canva.pillars.

Stubbed adapters (contributions welcome)

nano_banana -- Gemini 3 Pro Image via the nano-banana skill. Stub in image_generators/nano_banana.py. Implement generate() by calling the skill's image endpoint with a prompt built from the card spec.

midjourney -- Midjourney API (or proxy). Stub in image_generators/midjourney.py. Implement generate() by submitting a prompt, polling for completion, and downloading the result.

dalle -- OpenAI DALL-E. Stub in image_generators/dalle.py. Implement generate() using openai.images.generate.

Writing your own adapter

Subclass ImageGenerator from image_generators.base:

from image_generators.base import ImageGenerator

class MyAdapter(ImageGenerator):
    @property
    def name(self) -> str:
        return "my_adapter"

    def generate(self, spec: dict) -> dict:
        # spec keys: pillar, quote, handle, figure_path, output_path
        # return: {status: "ok", adapter: ..., image_path: ..., width: ..., height: ...}
        ...

Register it in image_generators/__init__.py and add a config block under image_generator.my_adapter in config.json.

Visual publishing pipeline

The visual_helper.py script handles the deterministic side of the visual queue pipeline: parsing a Review Queue markdown file, rotating through figures, and marking entries published. See visual_playbook.md for the full end-to-end flow including image generation, upload, and publishing steps.

python3 visual_helper.py peek --lang es        # get next approved visual item
python3 visual_helper.py mark --lang es ...    # mark published + update log
python3 visual_helper.py rotate-figure --pillar P1

Known gotchas

1. Cookie expiry. substack.sid cookies expire after a few weeks. Re-extract from Chrome DevTools when test_connection returns auth errors.

2. draft_bylines is required but undocumented. create_draft will 400 without it. The server fetches your user_id from /user/profile/self automatically and injects it into every draft POST.

3. Note attachment endpoint requires trailing slash. POST /comment/attachment/ (with slash) works. Without slash it returns 404. The server handles this correctly.

4. Publication-scoped vs global endpoints. Notes use substack.com/api/v1/comment/feed/. Post drafts use {subdomain}.substack.com/api/v1/drafts. Mixing them returns 404 or 403.

5. pledgedArr in the dashboard is a projection, not real pledges. Use get_earnings for actual revenue data.

6. Multi-publication support. Pass publication="name" to any tool to target a specific publication. Omit it to use default_publication from config.

7. Sections are required to publish on publications that have them. Substack returns HTTP 400: Please choose a section from publish_post / schedule_post when the draft hasn't been filed under a section. Look up section IDs with list_sections(publication="main"), then pass section_id=<id> to either the draft tool or the publish/schedule tool:

   code

   list_sections                             # → [{id: 12345, name: "Essays", slug: "essays", ...}, ...]
   create_draft title=... body=... section_id=12345
   # or, for an existing draft:
   publish_post draft_id=987654 section_id=12345
   

   publish_post / schedule_post patch the draft before publishing, so a single call is enough.

Related MCPs

Same author, same architecture pattern (FastMCP, draft+confirm on writes where applicable, vault auto-export, MIT):

• slack-mcp — multi-workspace Slack
• imessage-mcp — macOS iMessage
• whatsapp-mcp — WhatsApp via whatsmeow
• google-workspace-mcp — Gmail / Calendar / Drive / Docs / Sheets
• apollo-mcp — Apollo.io CRM + sequences
• luma-mcp — lu.ma events
• parse-mcp — markitdown / Docling / LlamaParse router
• rescuetime-mcp — RescueTime productivity data
• graph-query-mcp — vault knowledge graph queries
• graph-autotagger-mcp — wikilink suggestions from the graph
• investor-relations-mcp — seed-raise pipeline tracker
• vault-sync-mcp — bidirectional vault sync

Telemetry

This plugin sends a single anonymous install signal to myceliumai.co the first time it loads in a Claude Code session on a given machine.

What is sent:

• Plugin name (e.g. slack-mcp)
• Plugin version (e.g. 0.1.0)

What is NOT sent:

• No user identifiers, names, emails, tokens, or API keys
• No file paths, message content, or anything from your work
• No IP address is stored after dedup processing

Why: Helps the maintainer know which plugins people actually install, so attention goes to the ones that get used.

Opt out: Set the environment variable MYCELIUM_NO_PING=1 before launching Claude Code. The hook will skip the network call entirely. Already-pinged installs leave a sentinel at ~/.mycelium/onboarded-<plugin> — delete it if you want to reset state.

License

MIT

Built by Mycelium AI. Full install or team version at diazroa.com.