Connect an AI app
Connect an AI client to Neon Deer and choose what it can access.
Before you start
MCP lets an AI app use Neon Deer tools. You need an organization connected to Attio, access to the apps you want to use, and an alpha invitation for MCP and the relevant API families. Only an organization admin can create the connection.
The server address is:
https://api.neondeerdata.com/mcp
The alpha includes connection paths for Claude, Claude Code, ChatGPT, Codex, VS Code, Cursor, and Gemini CLI. Client availability can depend on the client's account and organization settings. Follow Settings → Integrations → Connect an AI app in Neon Deer for the current steps. Directory publication is not required to add the custom server.
Sign in from Claude Code
Add the remote server:
claude mcp add --transport http neon-deer https://api.neondeerdata.com/mcp
Run /mcp in Claude Code and sign in. In Neon Deer, check the requesting client, choose your organization, and review the permissions and resources before approving. Choose Off, Read, Read & write, or Custom for each offered app, and check the selected tables, objects, and workflows.
See Claude Code's MCP documentation for client-specific installation and authentication help.
Sign in from Codex
codex mcp add neon-deer --url https://api.neondeerdata.com/mcp
codex mcp login neon-deer
Complete the same organization and permission review in Neon Deer. See OpenAI's MCP documentation for client configuration.
Claude and ChatGPT
Add the server as a custom connector in Claude or a custom MCP app in ChatGPT, then sign in and review access in Neon Deer. A managed account may require its administrator to add or permit the connection first.
Follow Claude's custom-connector instructions or OpenAI's custom MCP connection instructions. Use the server address above; do not paste your Neon Deer key into a conversation.
VS Code, Cursor, and Gemini CLI
Add the server as an HTTP MCP connection in your client and follow its sign-in flow. The server setup page includes the current instructions for each client. Sign in to Neon Deer, choose the organization, and review app access and selected resources.
Use a key in Claude Code
If you use key-based authentication, select Create a key for Claude in Settings → Integrations. This key starts with no access. Choose the permissions, tables, objects, and workflows it needs and review them before creating it. Store the secret in the environment that launches Claude Code as NEON_DEER_API_KEY.
Add this entry under mcpServers in .mcp.json:
{
"mcpServers": {
"neon-deer": {
"type": "http",
"url": "https://api.neondeerdata.com/mcp",
"headers": {
"Authorization": "Bearer ${NEON_DEER_API_KEY}"
}
}
}
}
The file holds an environment-variable reference, not the secret. Run /mcp to check the connection. If a configured Authorization header is refused, fix that credential; the client does not necessarily fall back to sign-in.
Verify the connection
Start with a read-only prompt:
Which Neon Deer organization am I connected to, and what can this connection access?
Check the organization and permissions before making changes. Then try a resource the connection can read:
Show the fields in our Territories lookup table, then find the rows for Texas.
Show the latest company duplicate scan and its results.
Which scheduled workflow runs are due tomorrow? Use our organization's time zone.
Show recent email-alert delivery failures without including message contents.
These examples depend on the data and permissions you actually have. The action reference lists the available tools and links to their arguments, permissions, and effects.
Changes and permissions
MCP can make permitted writes, including row updates, scheduling, and authorized merges. It is not limited to reads. Grant access deliberately and inspect a proposed change before asking the client to perform it.
Permitted writes can take effect directly. Backup settings and permission to merge through the API or an AI app are changed in Neon Deer. Merge execution requires the organization's merge setting, the connection's merge permission, and the exact prepared plan revision. See permissions and merge plans.
For long-running work, ask the client to check the returned operation until it finishes and report individual failures. An accepted request is not a completed workflow or a successful merge.
Connect Attio for CRM values
Neon Deer's duplicate results contain record references rather than the records' CRM values. Add Attio's MCP connector when you want the assistant to retrieve those records and compare their details. Choose the same Attio workspace. Each connection has its own permissions.
Disconnect and troubleshoot
If you connected an AI app before v0.1.95, sign in again and review its organization, app access, and selected resources. Existing API keys are managed separately.
For a sign-in connection, open Settings → Integrations → Connected AI apps and choose Disconnect. For a key-based connection, revoke the key. Remove the server from your client afterward if you no longer use it.
Disconnecting does not undo completed writes or erase data already returned to the AI app. Privacy information covers Neon Deer's handling of customer data; the AI provider's terms also apply to information sent to that provider.
If access is refused, check the organization's alpha invitation and the connection's permissions. If a tool is missing, check its API-family access and required scopes. If a tool refuses a particular table, object, or workflow, check the selected resources too. If a write conflicts with a newer version, read the current resource and review it before retrying. Contact support@neondeerdata.com for help without sending keys or CRM data.
Read Why we built an MCP server for Neon Deer for the thinking behind the connector.