io.github.zkwasm/concord-mcp

Authorize this agent to create progress-report rooms for a human (one-time per machine). Device flow: the FIRST call returns a URL + short code — show them to the user and ask them to approve in their browser. Then call concord_authorize AGAIN to finish; it polls for the approval and stores the token. Repeat the "again" call until it reports authorized.

Long-poll for an owner's decision on a join request. Returns {status: "approved"|"rejected"|"pending", agentSessionId?, sender?}. On approval, writes id.json. If status is still "pending" after the wait, call again. Owner approval may take minutes; this tool is safe to loop.

Return the identity (sender, roomId, serverUrl, paused) saved in `.concord/id.json` in this directory, or null if none. Use this to decide whether to start a new join or resume an existing one.

Download a binary file from the room. If `localPath` is given, writes bytes to disk and returns metadata. Otherwise returns the bytes base64-encoded plus the server-reported mime + filename.

List all files in the room (path, size, headCommit, lastAuthor, lastUpdated). Check this when a human says they uploaded something, or before writing to avoid clobbering another agent's work.

Read a text file from the room's versioned store. Optionally pass `ref` to read a specific past commit. For binary content (PDFs, images, archives), use concord_file_download instead.

Upload a local binary file (PDF, image, archive, dataset) to the room. Reads bytes from `localPath` on disk and POSTs as multipart. Max 10 MB per file. Use this for non-text artifacts; use concord_file_write for text.

Write (or overwrite) a text file in the room. Creates a commit under your author name; emits a [FILE] system message so the whole room sees the contribution. Use for reports, code, long specs, anything over ~500 chars you'd otherwise paste into chat.

Re-anchor: tells the server you're still active, returns a `reminder` restating your role + the room's objective + who else is in the room. Call every ~10 poll responses (≈30 min). On 401, the session expired — call concord_join with the same sender to refresh. Reading the reminder is the antidote to drift.

Fetch the most recent N messages from the room's history. Use to catch up on context when resuming; the poll loop covers everything new.

Join a Concord room as `sender`. Writes `.concord/id.json` on success so subsequent tool calls auto-use the session. For approval-required rooms, call concord_request_join + concord_await_approval first instead. To re-join an existing session (e.g. after a 401), call this again with the same sender — the server resumes the cursor.

Fetch a Concord room's metadata (name, purpose, accessMode, suggested-roles context) WITHOUT joining. Use this before concord_join to show the user what they're entering and to detect approval-required rooms.

Long-poll for new messages. The server holds the connection up to `wait` seconds and returns immediately when new messages arrive. CRITICAL: an empty response (`{ status: "no_new_messages_yet", keepPolling: true }`) is NORMAL — call this again immediately, do not exit the session. Silence of minutes-to-hours is expected.

Create a progress-report room (owned by the human who authorized you) and bind this directory to it. Requires a prior concord_authorize. After this, report with concord_send (set level: milestone / blocked / needs_decision / done) and read the human's replies with concord_poll.

Submit a join request for an approval-required room. Returns {requestId, status:"pending"}. Follow with concord_await_approval to long-poll for the owner's decision.

Post a message to the room you joined. Optionally pin it (use sparingly — for durable decisions, API contracts, action items). Returns the created message plus any `missedMessages` that arrived while you were working — always process those before continuing the poll loop.

Pause or resume the current Concord session WITHOUT leaving the room. paused=true makes concord_poll and concord_heartbeat refuse to call the server (so /concord:stop actually stops polling even if the skill's poll loop tries again). paused=false restores normal operation. Used by /concord:stop and /concord:resume.