4Degrees AI Connector

A remote Model Context Protocol (MCP) server hosted at https://mcp.4degrees.ai/mcp. Once your firm enables it and a teammate authorizes their AI assistant, the assistant can pull contacts, deals, interactions, and network-strength signals from 4Degrees into the conversation — and, with the teammate’s explicit consent, take action on the CRM. Every write is permission-checked and audited; almost every write is undoable within ~5 minutes (contact reminders are the one exception). There are no destructive operations.

What it is #

The connector exposes three read tools over MCP: discover, query, and compare. All three are annotated readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false — so any MCP-aware assistant treats them as safe to invoke without a confirmation prompt.

Plus one write tool — execute_action — gated by per-user consent. Annotated readOnlyHint: false, destructiveHint: false (no deletes), idempotentHint: true. See Write actions for the consent flow and full list of supported actions.

Authorization uses OAuth 2.1 with Dynamic Client Registration (RFC 7591) and PKCE S256, so each AI assistant registers its own client and your firm doesn’t need to provision credentials by hand.

Enable the connector #

One-time, by an owner or admin of your firm’s 4Degrees workspace:

1. Sign in to 4Degrees.
2. Open Workspace Settings → Integrations.
3. Toggle “AI Connector” on. This grants every member the option to connect their own AI assistant. It does not connect anyone automatically.

Connect your AI client #

Pick the way you actually use AI today. Each path takes under five minutes; the OAuth handshake is the same.

Claude on the web (claude.ai) #

Best UX. No JSON to edit, no CLI required. Recommended for most deal-team users.

1. In claude.ai, click your profile avatar → Customize → Connectors.
2. Click + → Add custom connector.
3. Paste the MCP URL: https://mcp.4degrees.ai/mcp.
4. Click Add. Claude opens a new tab to 4Degrees — sign in and approve. You’re done.

ChatGPT (chatgpt.com) #

Now live in the ChatGPT App Directory. One-click install — no URL to paste.

Claude Code (CLI) #

For developers using Claude Code in the terminal.

claude mcp add --transport http 4degrees https://mcp.4degrees.ai/mcp

That’s it — on first tool use Claude Code opens a browser to complete OAuth. Verify with claude mcp list; the server shows up as 4degrees.

By default the connector is scoped to the current project. Add --scope user to make it available across all projects.

Claude Desktop (macOS / Windows app) #

The desktop app does not yet have a built-in remote-connector UI. Until it does, use the mcp-remote bridge below.

1. Install Node.js if you don’t already have it (the bridge runs via npx).
2. Open Claude Desktop → Settings → Developer → Edit Config. That opens claude_desktop_config.json.
3. Add the 4degrees entry inside mcpServers:

{
  "mcpServers": {
    "4degrees": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.4degrees.ai/mcp"]
    }
  }
}

4. Quit Claude Desktop fully and relaunch. mcp-remote opens a browser the first time so you can sign in to 4Degrees.

OpenAI Codex (CLI) #

For developers using the OpenAI Codex CLI.

Edit ~/.codex/config.toml (or .codex/config.toml in your project root) and add:

[mcp_servers.4degrees]
url = "https://mcp.4degrees.ai/mcp"

Run codex mcp login 4degrees to authorize, or Codex will prompt you on first tool use.

Tool reference #

Three tools, each annotated readOnlyHint: true. AI assistants compose them automatically — you typically don’t call them by hand.

curl -X POST https://mcp.4degrees.ai/mcp \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {"name": "discover", "arguments": {"resource": "actions"}}
  }'

Write actions #

The connector can now perform a fixed, enumerated set of write operations on your CRM — gated by an explicit per-user opt-in. Without consent, every write attempt is rejected before the dispatcher even resolves the target. The consent prompt is shown inline by the AI assistant the first time a write is requested; the user can revoke any time from app.4degrees.ai/settings/mcp.

What writes are supported

The live action registry is the source of truth — use discover with resource: "actions" to list every action with its payload schema. The shape of what's supported today:

• Contacts — create, rename, edit notes, edit social profiles, edit location, log interactions, add reminders, add/remove tags, edit custom fields.
• Deals — create, update, add notes, add associated contacts, move pipeline stage, add/remove tags, reassign owner.
• Companies — create, update, add notes, add/remove tags, reassign owner.

What writes are NOT supported

• No deletes. The connector cannot delete contacts, deals, companies, or notes. Cleanup is a manual operation in the web app.
• No bulk operations. Every action targets a single resource. The AI assistant must call the tool once per target.
• No cross-tenant writes. The dispatcher rejects any write targeting a resource not visible to the authenticated user.
• No background writes. Every action is triggered by an explicit user prompt in the AI assistant.

How the consent gate works

1. The AI assistant interprets the user’s prompt and decides a write is needed (e.g. "log a call with Sarah today").
2. The connector receives the execute_action call and checks the user’s write-consent flag.
3. If unset, the connector returns a consent-prompt envelope. The AI assistant renders that prompt back to the user inline.
4. The user clicks Allow in the prompt. The flag is set. The original write proceeds.
5. Subsequent writes proceed without re-prompting until the user revokes consent.

Auditing

Every successful write is recorded in the 4Degrees activity feed with attribution “via AI Connector”, the actor, the action key, and the prior-value snapshot (for undo). Admins can review the activity feed in the standard 4Degrees admin UI.

OAuth flow #

Each teammate authorizes their AI assistant separately. Tokens are issued per-seat, scoped read-only, and stored as SHA-256 hashes server-side — the original token value is never persisted. Connection setup uses OAuth 2.1 with Dynamic Client Registration so the AI assistant registers its own client.

Code samples #

Discovering OAuth metadata (any client)

# Step 1 — discover the auth server
curl -i https:/.4degrees.ai/.well-known/oauth-protected-resource

# Returns:
{
  "resource": "https://mcp.4degrees.ai/mcp",
  "authorization_servers": ["https://app.4degrees.ai"],
  "resource_documentation": "https://mcp.4degrees.ai/docs/connector",
  "scopes_supported": ["read"],
  "bearer_methods_supported": ["header"]
}

Calling a tool with a bearer token

# Once your AI client has a token, every tool call is JSON-RPC over HTTP.
curl -X POST https://mcp.4degrees.ai/mcp \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "compare",
      "arguments": {
        "resource": "deals",
        "metric": "count",
        "group_by": "stage",
        "filters": {"pipeline": "Enterprise"}
      }
    }
  }'

What data is accessible #

Read-only access to data your seat in 4Degrees already has permission to view, scoped to your firm:

• Contacts, companies, deals (and their custom fields, tags, owners, locations)
• Interaction metadata — subject, sender, date, type. Encrypted email and meeting bodies are never returned.
• Pipelines, stages, custom field definitions, team membership
• Portfolio dashboards (triage, warmest leads, news, job changes, reminders)
• Introduction paths — teammates with the strongest relationship to a contact or company

See the connector Privacy Policy for the full data inventory.

Security & control #

How to revoke access #

• From Claude: Settings → Connectors → 4Degrees → Disconnect.
• From ChatGPT: Apps → 4Degrees → Disconnect.
• From 4Degrees: revoke the access token from your account settings.
• From your firm’s administrator: disabling the AI Connector in workspace settings invalidates every token issued to org members.

Revocation is effective immediately on the next request — there’s no background sync to wind down.

Troubleshooting #

FAQ #

Last updated April 27, 2026. Found something inaccurate? support@4degrees.ai.