mdshare-mcp

The translation should be entirely i n English. Let me produce the final translation.

Need ensure all inline code and placeholders.

Let's craft now.# mdshare-mcp

MCP server that lets an AI chat (Claude, and other MCP-supporting clients) directly call a single publish_markdown tool to publish markdown content to an already-deployed mdshare service — without needing to download the file to your machine and then curl it yourself.

Architecture

Claude.ai (chat) → tool "publish_markdown" → mdshare-mcp (server này) → POST → mdshare → trả URL

This is an MCP server of the Remote (Streamable HTTP) kind, running independently on Render — different from a "local (stdio)" MCP server that only runs when installed on a personal machine. Because you use Claude from the web, the Remote HTTP version is the right choice: nothing needs to be installed on your machine.

Related MCP server: @nlspec/server

The 2 tools provided

• publish_markdown(content) — publishes markdown content to mdshare, returns a viewer URL and a raw URL.

• fetch_markdown(key_or_url) — fetches back the content of an already-published paste, based on a key or URL (accepting both domain/p/key and domain/p/key/raw). This is an important fallback tool: some AI chats (notably documented with ChatGPT) cannot fetch a URL that returns a plain Content-Type: text/plain — the built-in browser/tool of these clients is usually designed to read HTML pages (with <title>, the DOM structure, etc.), and may treat a plain text file as a "download file" and then refuse to read the content instead of displaying it. When you run into that situation, simply ask the AI chat to call fetch_markdown instead of opening the URL itself — the content will be fetched right on the server and returned directly in the tool result, without depending on the client's web-browsing ability.

Which platforms support custom MCP servers? (as of writing, subject to change)

Because this server strictly follows the MCP standard (Streamable HTTP), it will automatically work with any client that supports the standard — no per-platform code changes are needed. For the platforms marked "unclear", the most reliable approach is to check directly in that platform's Settings/Connectors section, as MCP support is changing very quickly among AI chat providers.

Step 1 — Deploy to Render

Exactly the same process as you did with mdshare:

1. Push the code (server.js, package.json, render.yaml) to a new GitHub repo, e.g. mdshare-mcp.

2. On Render: New → Blueprint → select the mdshare-mcp repo.

3. When asked for environment variables, enter:

   • MDSHARE_BASE_URL = the URL of the previously deployed mdshare service (for example https://mdshare-11s5.onrender.com — without a trailing /).

4. Click Apply.

Once deployment is done, Render gives you a URL in the form https://mdshare-mcp-xxxx.onrender.com. The actual MCP endpoint is at: https://mdshare-mcp-xxxx.onrender.com/mcp.

Step 2 — Connect to Claude.ai via a Custom Connector

1. Go to claude.ai → Settings → Connectors (or the equivalent item in Settings).

2. Select Add custom connector (or "Add more").

3. Enter the URL: https://mdshare-mcp-xxxx.onrender.com/mcp

4. Give it a memorable name, for example "mdshare".

5. Save it. Claude will automatically discover the publish_markdown tool.

After this step, in any conversation, you can turn on the "mdshare" connector (via the tools menu in the chat window), and then ask for example:

"Summarize the above content into markdown and publish it to mdshare for me"

Claude will automatically call the publish_markdown tool and return the result URL right in the chat.

Connecting from other AI chat apps (ChatGPT, etc.)

If that client supports MCP via Streamable HTTP (the standard MCP protocol, not Anthropic-specific), the process is similar: provide the URL https://mdshare-mcp-xxxx.onrender.com/mcp in that client's connector/tool configuration section. The exact syntax for adding a connector varies by platform — check the client's official documentation to find where the setting is.

Manual test (not through an AI chat; just to check the server works)

On Windows cmd:

curl -X POST https://mdshare-mcp-xxxx.onrender.com/mcp ^
  -H "Content-Type: application/json" ^
  -H "Accept: application/json, text/event-stream" ^
  -d "{\"jsonrpc\":\"2.0\",\"id\":1,\"method\":\"tools/list\",\"params\":{}}"

The result must contain the publish_markdown tool in the returned list.

Security notes

• This MCP server has no authentication — anyone who knows the /mcp URL can call the publish_markdown tool to post content to your mdshare. For personal use, the risk is low (the URL is not public and is hard to guess due to Render's random subdomain), but if you want to tighten it:

  • Add code to server.js that checks a secret header (e.g. X-Api-Key) before processing an MCP request.

• mdshare (the target service) originally has no auth by design, so this tool is simply automating the same manual action you used to do with curl — it does not add new risk on the mdshare side.

Design rationale

• Streamable HTTP rather than SSE (Server-Sent Events): SSE is the older-generation MCP protocol, and has been marked deprecated in the current MCP spec; Streamable HTTP is the current standard protocol, recommended by the official SDKs for remote MCP servers.

• Stateless (sessionIdGenerator: undefined): because the tool just does one simple operation (one POST, one output), there is no need to keep a session state across multiple calls — this keeps the server simpler and works well with serverless/free-tier environments that can restart a container at any time.