Nod for AI assistants

Your assistant can read the work. It cannot approve it.

Connect Claude, ChatGPT, Cursor, VS Code or Zed to your workspace. It sees the projects, the versions, the notes and the decisions, and it drafts the reply. The decision stays with a person, because a decision with a model's name on it is not a decision anyone can invoice against.

claude mcp add --transport http nod https://nod.byrewired.com/api/mcp

Three jobs

The parts of the week nobody wants

Chasing, summarising and checking scope are the three things an account manager retypes every Monday. All three are reading jobs, and reading is what an assistant is actually good at.

Chase the gate

Who is late, on what, for how long, and whether it is even worth chasing yet. Two days on a deck is not late; two days on a broadcast delivery is. It drafts the note; you send it.

Round-up for a client

What moved this week, written for the client rather than the agency, in English or Arabic. No gates, no version numbers, no internal names.

Is this in scope

A request against the agreed baseline, with the line that decides it quoted. If the baseline is silent it says so, because a guess there is what a dispute is made of.

It can read everything you can read. It can approve nothing, download nothing, and reach nothing you cannot.

  • There is no tool that approves, rejects or signs. Not behind a permission, not for an admin, not at all.
  • No tool, resource or prompt returns a frame, a render or a file. Delivery stays where it already is.
  • Every query runs as the person who connected it, under the same rules their browser gets.
  • Disconnecting takes effect in the same instant, not at the end of an hour.

What it can reach

50 tools, and not one of them approves anything

Each one is a question an agency actually asks, and each is bound to a permission you tick when you connect. Reading is the default; commenting, exporting and analytics are asked for separately.

  • nod_docketnod.read
  • nod_list_projectsnod.read
  • nod_project_assetsnod.read
  • nod_asset_versionsnod.read
  • nod_asset_commentsnod.read
  • nod_searchnod.read
  • nod_contractsnod.read
  • nod_retainer_ledgernod.read
  • nod_post_commentnod.comment
  • nod_start_exportnod.export
  • nod_export_statusnod.export
  • nod_analytics_waiting_on_menod.analytics
  • nod_analytics_waiting_on_clientnod.analytics
  • nod_analytics_money_outstandingnod.analytics
  • nod_analytics_money_agreed_vs_billednod.analytics
  • nod_analytics_retainer_burnnod.analytics
  • nod_analytics_approval_turnaroundnod.analytics
  • nod_analytics_overdue_windowsnod.analytics
  • nod_analytics_link_engagementnod.analytics
  • nod_analytics_activity_digestnod.analytics
  • nod_analytics_gate_ageingnod.analytics
  • nod_create_projectnod.write
  • nod_update_projectnod.write
  • nod_create_foldernod.write
  • nod_rename_foldernod.write
  • nod_move_foldernod.write
  • nod_upload_assetnod.write
  • nod_create_tasknod.write
  • nod_update_tasknod.write
  • nod_create_baselinenod.write
  • nod_add_baseline_linenod.write
  • nod_create_change_requestnod.write
  • nod_create_invoicenod.write
  • nod_add_invoice_linenod.write
  • nod_create_contractnod.write
  • nod_create_formnod.write
  • nod_add_form_fieldnod.write
  • nod_book_consumptionnod.write
  • nod_price_change_requestnod.write
  • nod_update_contract_factsnod.write
  • nod_resolve_commentnod.comment
  • nod_create_sectionnod.write
  • nod_open_retainer_periodnod.write
  • nod_close_retainer_periodnod.write
  • nod_send_invoicenod.send
  • nod_send_contractnod.send
  • nod_send_change_requestnod.send
  • nod_publish_formnod.send
  • nod_assign_formnod.send
  • nod_issue_review_linknod.send

Six apps

Whatever you already use

Two of them take a link and ask you for permission in the browser. The other four are configured by editing a file, so they use an API key. Both carry exactly the same permissions.

Claude CodeRun this in any terminal.Approved in your browser
Claude DesktopSettings, then Connectors, then Add custom connector.Approved in your browser
ChatGPTSettings, then Connectors, then add an MCP server.Approved in your browser
Cursor~/.cursor/mcp.json, or .cursor/mcp.json in a project.Configured with a key
VS Code.vscode/mcp.json in your project.Configured with a key
Zedsettings.json, under context_servers.Configured with a key

How the permission works

Not a key on a shared drive

Connecting opens a page in your browser that names the app, the account it will act as, the workspace, and every permission it is asking for. You tick what it gets. You can see it and cut it later.

  • It acts as one named person in one workspace, never as the workspace itself.
  • Permissions are ticked one at a time, and an assistant that needs another has to ask again.
  • Every call is logged with the tool, the account and the result.
  • OAuth 2.1 with PKCE, and the Model Context Protocol as written rather than as approximated.

One line

It takes about a minute

Sign in, open Settings, pick your app, and approve the connection. Nothing changes for anyone else in the workspace until you do.