Screvi MCP Server for Claude and AI Assistants

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to outside data. Screvi's MCP server gives your assistant access to your highlights, sources, and saved articles, so you can ask questions grounded in what you have read instead of what the model half-remembers.

Screvi's server lives at https://api.screvi.com/mcp. It works with any MCP client that supports remote servers with OAuth, including Claude (web, desktop, and mobile), Claude Code, Cursor, and ChatGPT.

MCP access requires an active Screvi subscription or trial.

What Your Assistant Can Do

Search your highlights by meaning, not just keywords. Filter by favorites, source type, source name, author, or tag.
Pull up a highlight or a source with its tags, notes, and every highlight from that book.
Read a saved article in full, with your highlights and notes on it, and ask questions about it.
Browse your tags and fetch the highlights under any of them.
Search your saved articles by title, author, domain, reading status, inbox or Later, favorites, tag, or whether you have highlighted them.
Resurface random highlights for a daily review, or run today's spaced-repetition review with the same set the app and the review email use.
List recent highlights by date, source, favorite, or whether they carry a note: "what did I highlight this week" needs no search query.
Save links and capture highlights (optional): save a URL to your inbox, store a passage the assistant found as a highlight, or add a note to one you already have.
Tag and tidy (optional): add or remove tags, archive, favorite, or move articles to Later, fix a bad title, or re-fetch an article that saved badly.

No tool can delete an article, source, or tag. Highlights can only be moved to the trash, where you restore them in one tap. Every other write is reversible from the app.

Connect to Claude

Go to claude.ai/settings/connectors and click Add custom connector.
Name it Screvi and set the MCP Server URL to https://api.screvi.com/mcp. Click Add.
Claude sends you to Screvi to authorize. Log in and click Approve. The consent screen lists exactly what you are granting.
Start a new conversation. Claude can now search your reading.

You can also open the connector from any chat with the + button, then Connectors. To disconnect, remove it from the same settings page, or from Settings > API > Connected AI apps in Screvi.

Connect to Other Clients

Any client that supports remote MCP servers with OAuth works the same way: add a server with the URL https://api.screvi.com/mcp, and complete the login when the client opens the Screvi authorization page.

Claude Code: claude mcp add --transport http screvi https://api.screvi.com/mcp, then run /mcp to log in.
Cursor: add the URL under Settings > MCP.
ChatGPT: add it as a connector in Settings > Connectors (developer mode).

Example Prompts

"What have I highlighted about decision-making?"
"Find my favorite highlights from Meditations by Marcus Aurelius."
"Show me articles I saved this month that I haven't started."
"Read the article I saved about deep work and summarize its argument."
"Save https://example.com/essay to my inbox and tag it 'writing'."
"Give me three random highlights to think about this morning."
"Which tags do I use most, and give me five highlights from the biggest one."
"Move everything I saved in the last three days to Later."
"Archive every newsletter in my inbox older than two weeks."
"Tag the last five highlights you showed me as 'stoicism'."
"This article saved as 'Untitled'. Fix the title."

Prompt Recipes

The server also publishes ready-made prompts. In clients that support them (Claude shows them as slash commands), pick one instead of typing the request:

Prompt	What it does
`triage-inbox`	Sorts your inbox into read now, later, and archive, then applies your choices.
`weekly-recap`	Groups what you saved and highlighted in the last N days by theme.
`daily-review`	Walks through today's review one highlight at a time with a reflection question each.
`quiz-me`	Tests your understanding of recent highlights or one book.
`book-review`	Writes a review of a book or article from your own highlights.
`find-connections`	Finds highlights from different sources that speak to one idea.

Available Tools

Read tools work with every connection.

Tool	What it does
`search`	Semantic and keyword search across all highlights. Filters: favorites, source type, source or author, tag.
`get-highlight`	One highlight with its source, tags, and notes.
`get-source`	One source (book, podcast, video) with a page of its highlights in reading order. Long books are paged. Also accepts an article ID.
`get-article`	The full text of a saved article, with your tags, note, and highlights on it. Long articles come in chunks.
`list-tags`	Every tag with highlight and article counts.
`get-highlights-by-tag`	Highlights under a tag, optionally a random sample.
`random-highlights`	A random sample of highlights, optionally limited to favorites, a type, a tag, or one source.
`search-articles`	Saved articles by title, author, domain, body text, reading status, inbox or Later, favorites, tag, save date, or highlight count.
`list-highlights`	Highlights in date order with a total, filtered by date range, source, favorite, note, type, or tag.
`daily-review`	Today's spaced-repetition review set, the same one the app and the review email show.

Write tools only work when the connection was approved with write access. Only delete-highlights takes anything away, and it only moves highlights to the trash.

Tool	What it does
`save-article`	Save a URL to your inbox, optionally with tags. The page is fetched in the background. Or pass the HTML you already have and it is stored without fetching.
`create-highlight`	Store a passage as a highlight, attached to a source or article if you like, with a note and tags.
`add-tags`	Tag highlights and/or articles. New tag names are created.
`remove-tags`	Remove tags from highlights and/or articles.
`update-highlight`	Add or change a note, favorite, or fix the text of an existing highlight.
`update-tag`	Rename or recolor a tag.
`archive-articles`	Archive articles, or restore them to the inbox.
`set-articles-later`	Move articles to Later, or back to the inbox.
`favorite-articles`	Favorite or unfavorite articles.
`update-article-details`	Fix an article's title or excerpt, or set your note on it.
`reparse-article`	Re-fetch an article from its URL when the saved copy is truncated or full of navigation text.
`mark-articles-read`	Mark articles read, or back to unread.
`delete-highlights`	Move highlights to the trash. Restorable from the app; the assistant should confirm first.

Read-Only or Read-Write?

Whether a connection can change anything is decided when you authorize it. Screvi's consent page shows what the client asked for: read access only, or read plus writing. Clients that were connected before write support existed stay read-only.

If your assistant reports "This connection is read-only", disconnect Screvi in the client and connect it again. Approve the new request if it includes write access.

Privacy and Security

OAuth 2.0 with PKCE. You log in on Screvi's own page. Your password never reaches the AI client.
Scoped access. Read and write are separate grants. Nothing can delete articles, sources, or tags; highlights can only be moved to the trash, where you restore them.
MCP tokens are not app logins. The token an AI client holds only works on the MCP server. It cannot sign in to the Screvi app, change your account, or create API keys.
Fetched on demand. Screvi returns only what a query asks for. Screvi does not store your conversations.
Revoke any time. Open Settings > API > Connected AI apps in Screvi to see every connected client and disconnect one. Access ends immediately, even if the client still holds its token. Removing the connector in the AI client works too.

Troubleshooting

The connector asks me to log in again. Connections refresh themselves in the background and only lapse after 90 days without use, or after you disconnect the app in Screvi. Log in once more and the connection resumes.

"Subscription required". MCP needs an active plan or trial. Check your plan on the pricing page.

Search returns nothing. New highlights are indexed within a few minutes of import. If the library is fresh, wait briefly and try again.

"Article body is still being fetched". Articles saved through MCP are parsed in the background. Ask again a few seconds later.

Public API for scripts and custom integrations
AI Semantic Search to see how search ranks results