Docs · AI agents

Use Noon from Claude, ChatGPT, and Claude Code

Noon runs a remote MCP server. Connect it once and your assistant can create roles from a job description, review the ranked candidate feed, accept or reject with reasons, and edit email and SMS outreach sequences, all inside your Noon workspace.

Setup

You need a Noon account and an MCP client. The server URL is in your Noon workspace under Settings → MCP; sign-in is OAuth, so there is no API key to copy.

Claude (web and desktop)

  1. Open Settings → Connectors and click Add custom connector.
  2. Name it Noon and paste the Noon MCP server URL from your Noon workspace (Settings → MCP).
  3. Sign in with your Noon account when prompted and approve access.
  4. In any chat, click + → Connectors and enable Noon.

ChatGPT

  1. Open Settings → Apps and turn on Developer mode under Advanced settings.
  2. Create a new app, paste the Noon MCP server URL, and choose OAuth.
  3. Sign in with your Noon account, then enable Noon from the + menu in a chat.

Claude Code

  1. Run `claude mcp add -s user -t http noon <your Noon MCP server URL>`.
  2. Run `/mcp` inside Claude Code and complete the sign-in.

Other MCP clients (Cursor, Windsurf, custom agents)

  1. Add a remote MCP server with transport “Streamable HTTP” and the Noon MCP server URL.
  2. Authentication is OAuth 2.1 with PKCE and dynamic client registration, so no API key is needed; the client opens a Noon sign-in.

What the assistant can do

Each tool maps to something a recruiter does in Noon. Names below are the tool names your client will show.

Roles and sourcing

create_roleStart sourcing from a pasted job description or a job-posting URL. Noon extracts the title, location, and non-negotiables and begins its search.
list_rolesList the roles in the workspace with their status.
get_role_stateRead a role's current search settings: titles, locations, must-haves, and what has been relaxed.
find_candidatesRun a one-off candidate search with explicit criteria, without creating a role.

Reviewing the feed

get_candidate_feedPull the next batch of ranked candidates for a role, with Noon's reasoning for each.
get_candidateFetch one candidate's full profile and evaluation.
explain_candidateAsk why a candidate was ranked where they were, criterion by criterion.
accept_candidate / reject_candidateAccept into the project or reject with a reason. Rejections teach the search what to avoid.
give_feed_feedbackGive feed-level feedback in plain English (“too many agency recruiters”, “prefer Series A–C companies”).
list_rejection_reasons / edit_rejection_reason / unlearn_rejection_reason / undo_rejectionInspect and correct what Noon has learned from past rejections.

Outreach

get_outbound_sequenceRead the outreach sequence for a role: steps, wait times, and message copy.
update_outbound_sequenceRewrite steps or timing. Email and SMS steps send from the senders connected in Noon.

Diagnostics

pool_diagnosticsSee how large the candidate pool is for the current criteria and which criterion is the bottleneck.
rejection_analysisSummarise why candidates are being rejected and what to loosen.
search_trace / query_eventsTrace what the search did and when; useful when a role looks thin or slow.

A typical session

  1. “Source a Staff Backend Engineer in Austin from this JD.” The assistant calls create_role; Noon starts searching immediately.
  2. “Show me the top 10 and why.” get_candidate_feed plus explain_candidate for anyone you ask about.
  3. “Reject the agency recruiters, accept the rest, and prefer people who have shipped on Go.” reject_candidate, accept_candidate, give_feed_feedback; the search re-ranks.
  4. “Why is the pool thin?” pool_diagnostics names the criterion to loosen.
  5. “Shorten the second email and move it to day 3.” get_outbound_sequence then update_outbound_sequence.

For agent instruction files (AGENTS.md, CLAUDE.md, .cursorrules)

Paste this into the instructions your agent reads so it knows when and how to use Noon.

## Noon (AI recruiter) via MCP
- Noon sources, ranks, and contacts job candidates. Use its MCP tools when the user wants to hire or source for a role.
- Start with create_role (paste the JD) or find_candidates (explicit criteria). Then get_candidate_feed and explain_candidate to review.
- Record decisions with accept_candidate / reject_candidate so the search learns; use give_feed_feedback for pattern-level feedback.
- Use pool_diagnostics or rejection_analysis before loosening criteria. Read get_outbound_sequence before editing outreach.
- Full docs: https://www.noon.ai/docs/mcp  Plain-text site index: https://www.noon.ai/llms.txt

Machine-readable versions of this site: /llms.txt (every page with a summary) and /llms-full.txt (full text). Any public page also returns markdown when requested with Accept: text/markdown.

FAQ

What is MCP?

The Model Context Protocol is an open standard that lets an AI assistant call tools on another product. Noon runs a remote MCP server, so Claude, ChatGPT, Claude Code, and other MCP clients can source, review, and manage outreach in your Noon workspace from a chat.

Where do I find the server URL?

In your Noon workspace under Settings → MCP. It is tied to your account through OAuth, so sharing the URL does not share access to your data.

Do I need a separate plan?

The connector is part of Noon; it uses the same account, roles, and senders as the web app. Ask us for current availability details for your team.

What can the assistant not do?

It cannot connect email or SMS senders, change billing, or invite teammates. Those stay in the Noon web app. Every send still goes through the senders and limits you configured there.

Which outreach channels does Noon use?

Email and SMS. Sequences are written and scheduled by Noon and sent from the inboxes and numbers connected in your workspace.

Weekly product changes are on the changelog.