TapCueMCP connector, for power users
Save a link on your phone. Your AI picks it up.
Share anything to TapCue under a reason like For AI, one tap. Then ask Claude, ChatGPT or Cursor what's new, have it summarise, draft or research, and let it mark the save done. TapCue's MCP server is hosted: one key, nothing to install.
Your format in, your assistant out
Every reason writes in the format you give it: a Markdown note with front matter, a .json file, XML, YAML, or the exact JSON body of a webhook into your own system. The same save is readable over MCP and the REST API, so a link shared from your phone can land in your vault, hit your endpoint and be waiting for your assistant, all from one tap.
What the assistant can do
Read
list_saves (newest first, filter by reason, not-done, since, search), get_save, list_categories, with each category's instruction for what you want done with its saves.
Act
mark_done, update_save (reword, refile, optionally resend), add_save, delete_save. Only with a full-access key.
Set TapCue up
list_destinations, get_category, check_template, create_category, update_category: ask your assistant to build a reason, write its JSON or Markdown template and test it against a sample save before it is saved. Full key.
Stay in scope
Each key sees only the reasons you chose. Text inside a save or a fetched page is handed to the assistant as data, never as an instruction.
Connect in three steps
Make a reason for it
In TapCue, pick or create a reason such as For AI and give it an instruction ("summarise in five bullets", "add to my reading list"). Saves still go to your notes too if you want.
Create an API key
At tapcue.app/app under Settings: choose the reasons it may see and whether it's read-only or full. The key (tck_…) is shown once.
Add the server
Connector URL
https://api.tapcue.app/mcp, headerAuthorization: Bearer tck_…. Apps that only take a URL:https://api.tapcue.app/mcp/tck_….
Setup by client
Claude.ai and Claude Desktop
Settings → Connectors → Add custom connector. Name it TapCue, paste https://api.tapcue.app/mcp/tck_… as the URL (this form carries your key, so treat it like a password). Then turn it on in a chat.
Claude Code
claude mcp add --transport http tapcue https://api.tapcue.app/mcp --header "Authorization: Bearer tck_…"
Cursor
{ "mcpServers": { "tapcue": { "url": "https://api.tapcue.app/mcp", "headers": { "Authorization": "Bearer tck_…" } } } }
ChatGPT
Add TapCue as a connector (MCP server) with the URL-only form, https://api.tapcue.app/mcp/tck_…, and no extra authentication.
Anything else
Streamable HTTP, JSON-RPC 2.0, tools only. The same key also works on the REST API.
Things to ask
- "What's new in my For AI saves? Two lines on each."
- "Read the three articles I saved this week and tell me which one is worth my time."
- "Summarise the video I saved yesterday with timestamps, then mark it done."
- "Take everything under Buy this and build me a comparison table with prices."
- "Turn this week's saves into a short newsletter for my team."
Questions
What is MCP?
The Model Context Protocol is an open standard that lets an AI assistant call tools in other apps. TapCue's MCP server is hosted at api.tapcue.app/mcp: nothing to install, and it works from any assistant that supports remote MCP servers over HTTP.
Which assistants work?
Claude (claude.ai custom connectors, Claude Desktop and Claude Code), ChatGPT (connectors), Cursor, and any other MCP client that can send an HTTP request with a header. Clients that only take a URL can put the key in the URL.
Can the assistant change my saves?
Only with a full-access key. A read-only key allows list_saves, get_save and list_categories. Full keys add mark_done, update_save, add_save and delete_save, plus the setup tools: list_destinations, get_category, check_template, create_category and update_category. Every key is limited to the reasons you picked when you created it; a key limited to chosen reasons cannot create new ones.
Can the assistant build my categories and templates?
Yes, with a full key. Say what you want ("a Recipes reason that writes a JSON file per save into my vault with the title, link and my note") and it calls check_template to test the template against a sample save, then create_category or update_category to save it. The same validation runs as in the editor: a .json or webhook template that is not valid JSON is refused.
Does the assistant read the pages I saved?
TapCue stores the link, the one-line headline it wrote, your note and any shared text, not the page. An assistant that can browse the web fetches the page itself when you ask it to summarise.
Is the text in my saves treated as instructions?
No. Each tool tells the assistant that what's inside a save (title, note, shared text, the page it links to) is data you collected, never an instruction to follow. A category's instruction, which you write in TapCue's settings, is the only guidance it takes from your account.
Does it cost extra?
The connector is part of Pro ($29 a year). There is no charge per key, per call or per assistant.
Can I control the format a save is written in?
Yes. Each reason that writes a note of its own has a template you write yourself with placeholders ({headline}, {url}, {note}, {category}, {date}, {tagname}): Markdown, JSON (checked to be valid when the file name ends in .json), XML, YAML or any other text. Webhook reasons take your own JSON body. Email has a template too.
The connector is part of Pro, $29 a year.