Blog Post MCP (Archy)

Minimal MCP server that exposes tools to:

• summarize the current git workspace changes
• create/update/publish blog posts
• trigger featured image / featured video generation via the existing Archy admin API

Env vars

• ARCHY_BLOG_BASE_URL (default: http://localhost:4280)
• ARCHY_PUBLISHER_KEY (recommended; per-user key sent as x-archy-publisher-key)
• ARCHY_INTEGRATIONS_API_KEY (legacy/shared integration key; sent as Bearer or x-archy-api-key)
• ARCHY_API_KEY_HEADER (optional: authorization, x-archy-api-key, or x-archy-publisher-key; default: authorization)

Auth notes

• In your API, server-to-server auth is controlled by INTEGRATIONS_API_KEY + INTEGRATIONS_API_ROLES.
  • Publishing and AI image/video require admin role.
  • Draft creation can work with writer role.
• For local development you can bypass auth entirely by setting LOCAL_DEV_ADMIN_BYPASS=true in the API environment.

Run

Standalone (recommended)

After publishing this package, you can run it without cloning meet.archy-ai.com:

ARCHY_BLOG_BASE_URL=https://meet.archy-ai.com \
ARCHY_INTEGRATIONS_API_KEY=... \
npx -y @archy-ai/blog-post-mcp

Notes:

• The archy_workspace_summary tool is optional; it will only be useful when run in a folder that is a git repo.

CLI flags (no env vars)

The published binary also supports flags that populate env vars internally (useful for VS Code MCP configs):

npx -y @archy-ai/blog-post-mcp \
	--base https://meet.archy-ai.com \
	--header x-archy-publisher-key \
	--publisher-key pk_...
	--timeout-ms 120000 \
	--debug

Notes:

• --debug logs only to stderr (stdout stays clean for MCP protocol).
• --timeout-ms controls how long the MCP waits for API responses before aborting (useful when image/video prompts are complex).

From source

cd meet.archy-ai.com/tools/blog-post-mcp
npm install
ARCHY_BLOG_BASE_URL=http://localhost:4280 \
ARCHY_INTEGRATIONS_API_KEY=... \
node ./index.mjs

You can also pass flags when running from source:

node ./index.mjs \
	--base http://localhost:7072 \
	--header x-archy-publisher-key \
	--publisher-key pk_...

Smoke test (recommended)

Use this when you want a repeatable, one-command verification that:

• VS Code can spawn the MCP server (stdio)
• the full toolchain works end-to-end
• the backing local SWA dev server is reachable (ARCHY_BLOG_BASE_URL)

What it does

The smoke script runs only via MCP tool calls (not curl):

1. archy_blog_create_draft (creates a dev-only draft)
2. archy_blog_ai_draft (fills body + SEO)
3. archy_blog_ai_image (generates a featured image)
4. archy_blog_attach_feature_image
5. archy_blog_ai_video (starts a featured video job conditioned on the image)
6. archy_blog_ai_video_status (polls until succeeded/failed)
7. archy_blog_attach_feature_video
8. archy_blog_publish

Prereqs

• Start your local SWA stack so http://localhost:4280 is serving:
  • npm run swa:dev (from meet.archy-ai.com)
• Auth:
  • Local dev: set LOCAL_DEV_ADMIN_BYPASS=true in the API environment (recommended), OR
  • Provide ARCHY_INTEGRATIONS_API_KEY / INTEGRATIONS_API_KEY so the admin endpoints authorize.

Run

From meet.archy-ai.com:

npm run mcp:blog:smoke

Optional flags:

node ./scripts/archy-blog-mcp-smoke.mjs \
	--base http://localhost:4280 \
	--title "MCP Smoke Test (manual title)" \
	--video-polls 30 \
	--video-poll-ms 5000

Expected result

On success, it prints a single JSON object like:

{
	"ok": true,
	"baseUrl": "http://localhost:4280",
	"post": {
		"id": "...",
		"slug": "...",
		"status": "published",
		"publishedAt": "..."
	},
	"featureImage": { "assetId": "...", "url": "..." },
	"featureVideo": { "requestId": "...", "assetId": "...", "url": "..." }
}

If it fails, it exits non-zero and prints an error with context.

Agentic composer

These tools help an agent compose a validated TipTap/ProseMirror doc body from a safe “blocks” spec, then write it to a post.

Tools

• archy_blog_get_post: fetch a post by id
• archy_blog_compose_doc: convert blocks → TipTap doc JSON (no network)
• archy_blog_set_body_from_blocks: replace the post body with blocks
• archy_blog_append_blocks: append blocks to an existing TipTap doc

Block format (overview)

Each block is an object with a type.

Supported type values:

• heading (attrs: level 1-3; content from text or inlines)
• paragraph (content from text or inlines)
• quote (paragraphs: string[])
• bulletList, orderedList (items: string[])
• image (src, optional alt)
• video (src)
• embed (url, optional provider)
• slider (items: [{ src, alt?, assetId? }])
• card
• cardSet (cards: [...])
• cta

Inline rich text is supported in heading and paragraph via:

inlines: [{ text, bold?, italic?, href? }] where href must be http(s)://... or a site-relative path like /pricing.