alden

Alden in your AI coding tool

alden mcp gives an AI coding tool (Claude Code, Cursor, Claude Desktop, or anything else that speaks the Model Context Protocol) Alden's queue, briefings and memory as tools. Ask "what's waiting on me?" or "review PR 3 the way I would" and the assistant uses Alden's checks, the callers it finds outside the diff, and what Alden knows you look for, then adds its own reading of the code.

It's the same Alden as the terminal and alden ui: your GitHub sign-in, your settings and your memory, all on your machine.

Add it

Claude Code, in a clone of a repo you review (Alden finds callers in the clone the tool starts it in):

claude mcp add alden -- alden mcp

Add --scope user to have it in every project. For the slash command /alden:review as well, install the Claude Code plugin instead: see Claude Code plugin.

Cursor: in .cursor/mcp.json (or ~/.cursor/mcp.json for every project):

{ "mcpServers": { "alden": { "command": "alden", "args": ["mcp"] } } }

Claude Desktop: in Settings → Developer → Edit Config, add the same alden entry under mcpServers. Apps opened from the Finder may not find alden on your PATH, so give its full path (which alden prints it, e.g. /opt/homebrew/bin/alden). Claude Desktop has no project folder, so point Alden at a clone for callers and local changes: "args": ["mcp", "--dir", "/path/to/your/clone"].

Sign in to GitHub first (alden auth login); the tools say so if you haven't.

The tools

Tool What it does Arguments (default)
alden_queue The PRs waiting on your review, ranked, each with why it's there. repos, orgs, limit 1–100 per lane (15), refresh (false)
alden_review A briefing on a PR (a URL, owner/repo#123, or its number in the queue), or on your uncommitted changes when no PR is given. Includes what you care about in that repo. pr (none: your local changes), callers (true), use_alden_model (false), model, effort low to max
alden_feedback Records useful, not useful or dismiss on a place in a briefing, by its id (E1, F2, C:tests.missing), and says how close it brought a lesson. place, verdict useful, not_useful, dismiss or clear, pr (the latest briefing)
alden_missed Records something Alden should have flagged. note (up to 2,000 characters), path, line, pr (the latest briefing)
alden_memory What Alden knows about you in a repo. repo (this clone's origin)
alden_remember Adds a note to it ("Money is in cents, never floats"). text (up to 500 characters), repo (this clone's origin)
alden_draft_comment Drafts a review comment for you, on a line or lines of the PR's diff or as a general note, or a suggested change you can apply on GitHub. On code outside the diff (a line GitHub won't take an inline comment on, or a whole file) it drafts a linked note, a general note in the review body that links to that code, and tells the assistant so. With in_reply_to, it drafts a reply in a review thread already on the PR instead. It isn't posted: you review it and send it from the UI or with alden send. pr, path and line (none: a general note; path alone: a note linking to the whole file), start_line (for several lines, in one hunk), side RIGHT or LEFT (RIGHT), body (optional with a suggestion), suggestion (replacement code for the line(s), new side only; empty deletes them; in a linked note, code to copy), in_reply_to (a thread's file:line or id from alden comment <pr> --threads; not with path, line or suggestion)

Each place in a briefing lists the files to look at too, when the code graph ran: callers of the changed code, tests that use it, and files importing it, each with its line (see Related files).

The assistant records feedback when you react ("that first one wasn't useful"), so Alden keeps learning what you look for wherever you review. It's told to record only what you said, never its own opinion.

Ask it to write a comment, or to suggest a change, and it drafts one in Alden with alden_draft_comment, marked as drafted by Claude Code (or whichever tool you use). Asked to comment on a caller outside the diff, it gets a linked note back, and the tool's answer says so. Asked to answer a comment on the PR, it drafts a reply in that thread (in_reply_to), which goes with your review like any draft. Alden posts only what you send from the UI or terminal: the assistant can draft comments and replies, you review and send. There is deliberately no tool to send, approve, request changes or resolve a thread. See Comments and sending a review.

The model

By default alden_review doesn't ask the model you set up for Alden: the assistant you're talking to is already a model, so it reasons over Alden's checks and the code graph and reads the diff itself, and Alden costs nothing extra. Ask for "Alden's model too" and it passes use_alden_model, which uses your Alden model within your caps, as alden review does.

Safety