Skip to main content
Docs

Connect Bucky to your AI

Add Bucky to Claude, Claude Code, Cursor, VS Code, Codex, ChatGPT, Grok, or the CLI.

Bucky runs an MCP server, so an AI assistant can look up zoning, parcels, and feasibility for a street address without you leaving the conversation.

You need one thing:

https://buckybuild.com/api/mcp

Everything below is a different way of handing that URL to a client. Pick yours.

Claude

Claude web and Claude Desktop both support remote MCP servers as custom connectors. There is no install link — Anthropic does not offer one — so this is a paste.

  1. Open Settings → Connectors.
  2. Choose Add custom connector.
  3. Paste https://buckybuild.com/api/mcp and save.
  4. Ask Claude something that needs Bucky — "what can I build at 412 Main St?" — and approve the sign-in page it opens.

Custom connectors need a paid Claude plan (Pro, Max, or Team). If you are on the free plan, use Cursor or the CLI below instead.

Cursor

Cursor supports one-click install:

Add Bucky to Cursor

Or add it by hand in Settings → MCP → Add new MCP server, pasting this object. Cursor registers itself automatically; you do not need to create anything on the Bucky side first.

{ "type": "http", "url": "https://buckybuild.com/api/mcp" }

The manual route is not a worse route. A one-click install dialog shows you a command in a single-line field, and Cursor's has a known weakness where a long enough command can scroll the dangerous part out of view — which is a good reason to prefer reading the config yourself, for any server, including ours. The JSON above is exactly what the button writes.

VS Code

Add Bucky to VS Code

Or add the same object by hand, in .vscode/mcp.json for one workspace or through MCP: Add Server for all of them:

{ "type": "http", "url": "https://buckybuild.com/api/mcp" }

Claude Code

claude mcp add --transport http bucky https://buckybuild.com/api/mcp

Claude Code opens the sign-in page the first time a tool needs your account.

Codex

Codex reads ~/.codex/config.toml. A url key rather than a command is what selects the Streamable HTTP transport:

[mcp_servers.bucky]
url = "https://buckybuild.com/api/mcp"

Then run the OAuth leg once:

codex mcp login bucky

Or let the CLI write the entry and run that OAuth command:

bucky mcp install codex

ChatGPT

Read the plan gate before the instructions, because on ChatGPT it decides whether any of this works:

  • Developer Mode — the route below — needs a Business, Enterprise, or Edu plan.
  • Pro can use connectors for read and fetch actions only.
  • Free cannot add custom connectors at all.

The route with no plan gate is the Apps Directory: one click, OAuth handled, no URL to paste and no developer mode. Bucky's directory listing is with OpenAI for review; this page will link it here when it is published, and that will be the recommended path for everyone.

Until then, on a plan that allows it: open Settings → Connectors, turn on developer mode, add https://buckybuild.com/api/mcp, and approve the sign-in page. Note that the developer mode toggle has moved between ChatGPT builds — if you cannot find it, that is the more likely reason, not your plan.

If you are on Free or Pro, use Claude, Cursor, VS Code, or the CLI below today.

Grok

Open grok.com/connectors, choose New Connector, then Custom, and paste https://buckybuild.com/api/mcp. Grok reads the tool list from the server. Custom connectors are on Grok's paid tiers.

Grok Build, the terminal CLI, reads MCP configuration you already have — nothing extra to do there.

Windsurf, Zed, Cline, and everything else

Any client that speaks Streamable HTTP takes the same object in its own MCP config file:

{ "type": "http", "url": "https://buckybuild.com/api/mcp" }

Or let the CLI write it:

bucky mcp install windsurf

bucky mcp install also writes Claude Desktop, Zed, Cline, Cursor, VS Code, and Claude Code. Pass no argument and it offers the clients it finds on this machine. --dry-run prints the diff.

Bucky CLI

bucky auth login

This opens a browser, signs you in, and stores the token locally. No connector configuration needed.

Signing in

The first time your assistant calls a tool that needs your account, it opens a Bucky sign-in page. You will see:

  1. Which app is connecting — named at the top of the sign-in form.
  2. What it will be able to do — the exact permissions, listed before you approve. Reading your projects, creating and updating them, and exporting report PDFs are separate permissions, because exporting spends your export credits.

Approve once and the connection persists. You never paste a token, and the assistant never sees your password.

Two tools work with no account at all, so you can check the connection is live before signing in: getBuckyMcpGuide and explainFeasibilityStatus.

Managing and disconnecting

Connected apps are listed at Settings → Connected apps in Bucky, with what each one can do and when it was last used. Disconnecting takes effect immediately — the next request that app makes is refused, rather than working until its token happens to expire.

If it does not connect

  • "Invalid client" — your client registered but could not be resolved. This was a bug fixed on 2026-08-17; make sure the client is not caching an old registration, and try removing and re-adding the connector.
  • Signed in but nothing happened — the connector flow needs to return you to Bucky's authorization page after login. If you land on the Bucky dashboard instead, the flow was interrupted; start it again from your assistant.
  • "New Bucky accounts are paused" — self-serve signup is currently closed. Join the waitlist, and connect once your account exists.

For developers

Protocol details — transport, discovery documents, verifying the endpoint with curl — are in MCP integrations. The tool catalog is in MCP tools.

Last updated on