Reference

Connect Wesley to Claude or ChatGPT

Copy a setup prompt, connect your AI app, and start working with your Wesley accounting data. No installation or coding needed for the web apps.

Ask your AI app to work with your Wesley data: review transactions, explain a report, or help with bookkeeping. Wesley's remote MCP connection uses your personal API key and your current permissions.

Start by copying a prompt

Copy this setup prompt and paste it into Claude or ChatGPT. Your AI app can guide you through setup. You still add the connection and approve it yourself; a prompt cannot grant access to your account.

Use Wesley in Claude or ChatGPT

Copy the setup prompt and paste it into your AI app. It will guide you through connecting Wesley. You approve the connection and enter your API key on Wesley.

Your API key is never included in the copied prompt.

MCP server URL

https://www.wesley-ai.co/api/mcp
View setup prompt
Help me connect Wesley to this AI app so I can work with my accounting data.

Official setup guide: https://www.wesley-ai.co/docs/reference/mcp
Remote MCP server: https://www.wesley-ai.co/api/mcp

If Wesley tools are already connected, call wesley_api_read with path /api/users/me, then help me choose my firm and company.

Otherwise, read the setup guide and walk me through the connection steps for this app. In Claude or ChatGPT, I need to add the remote MCP server and approve the connection myself. Use OAuth when offered. I will enter my personal API key only on Wesley's connection page. If this app supports configuring MCP with an Authorization header, use the WESLEY_API_KEY environment variable or its secret settings.

Never ask me to paste an API key into chat, put a key in a URL, or claim the connection is complete before a Wesley tool call succeeds. If this app cannot install or use MCP, explain that and direct me to a supported app.

After connecting, use wesley_api_catalog to discover the relevant API. Ask which company I mean if it is ambiguous. Make changes only when I request them.

This guide is public and can be shared without signing in. The copied prompt contains setup instructions and the server URL, never your key.

Connect in three steps

  1. Create a Wesley key. In Wesley, select your firm and open API Keys & MCP → Create key. Use a name such as “Claude” or “ChatGPT.” Copy the key when it is shown. Existing keys also work if you still have their value. Personal API access requires a paid firm.
  2. Add Wesley to your AI app. Use the server URL below and choose OAuth when asked about authentication. The app-specific steps are below.
  3. Approve on Wesley. The connection opens a Wesley page. Check the app and return address, paste your key there, and select Connect to Wesley. Return to your AI app and ask it to check the connection.
https://www.wesley-ai.co/api/mcp

Enter your key only on Wesley's connection page. Do not paste it into an AI conversation or add it to the server URL. The AI app receives a separate connection token; Wesley stores the underlying key encrypted for the connection. Revoking the key stops access, including existing MCP connections.

Claude

Open Claude's Settings → Connectors, add a custom connector named Wesley, and enter the server URL. Connect it and complete the Wesley approval page. Use Wesley in your conversation after connecting.

Leave optional OAuth client ID and client secret fields empty: Wesley supports automatic client registration. Workspace administrators may control which connectors can be added. See Claude's custom connector guide.

ChatGPT

Enable Developer mode in ChatGPT's settings, then create an app for the remote MCP server from ChatGPT Plugins. Name it Wesley, enter the server URL, and choose OAuth. Use dynamic client registration (DCR) if an OAuth registration option is shown; leave optional static client credentials empty. Complete the Wesley approval page and select Wesley for your conversation.

Developer mode availability and workspace controls depend on your account. Follow the current ChatGPT developer mode instructions if your settings look different. The ordinary GPT Builder Actions editor is a separate OpenAPI integration; use an MCP app for this connection.

Try your first request

Use Wesley to check which account is connected and list the companies I can access. Ask me which company to work on before reading its accounting data.

Then try a request such as:

Show this company's profit and loss for last month. Explain the largest changes from the previous month. Do not change any records.

The connection supports both reading and changes through the APIs your key can access. Ask explicitly when you want changes. Existing read-only roles remain read-only, and actions that need approval inside Wesley still need that approval.

Other MCP clients and developer APIs

Clients that support custom request headers can use the personal API key directly, without the browser approval flow:

SettingValue
TransportStreamable HTTP
Server URLhttps://www.wesley-ai.co/api/mcp
HeaderAuthorization: Bearer <your Wesley API key>

Store the key in the client's secret configuration or WESLEY_API_KEY environment variable. Environment variable substitution depends on the client; do not assume a literal variable name in a JSON configuration will expand automatically. OpenAI's Responses API supports supplying a remote MCP bearer token in its authorization field; see OpenAI's MCP documentation.

For agents: use the connection

  1. Call wesley_api_read with path: "/api/users/me" to verify the account.
  2. Read /api/users/accounting-firms, then /api/users/companies with the selected accountingFirmId query parameter. Ask the user if the target is ambiguous.
  3. Search wesley_api_catalog for the workflow. Pass an exact returned path template to inspect its full method-specific contract, including required and optional fields.
  4. Use wesley_api_read for GET requests or wesley_api_write for POST, PUT, PATCH, and DELETE. Supply company_id and accounting_firm_id when required.
  5. Use pagination and returned IDs. Respect http_status, retry_after, and errors. Do not blindly repeat writes after timeouts or uncertain outcomes.

The server exposes key-compatible APIs from Wesley's API catalog. Each API enforces the key owner's current permissions, paid access, roles, quotas, and action approvals. MCP checks required context and documented inputs before forwarding the call. Database schema inspection and arbitrary SQL through /api/agent/read-query are not exposed; other APIs are available only when their own handler supports the key and authorizes its owner. Browser-only flows still need the user to finish them in Wesley.

JSON bodies and multipart fields/files are supported. MCP accepts requests up to 3 MiB, with at most 2 MiB of decoded files, and returns API response bodies up to 1 MiB. Use smaller pages for reads and the direct API for larger file transfers. A large response may be omitted after a write succeeds; inspect its status before retrying.

Read the personal API key guide for API examples, quotas and approval requirements. A Markdown version of this setup guide is available at /llms.mdx/docs/reference/mcp.

If connecting does not work

  • No connector option: Check whether your AI app/account supports custom MCP connections and whether your workspace administrator allows them. A pasted prompt cannot enable that feature.
  • Invalid key or access denied: Create a new key in Wesley, confirm paid-firm access, and connect again. Expired/revoked keys and disabled accounts cannot authenticate.
  • Connection expired: Browser connections last at most 30 days, and can end sooner if the key expires, is revoked, or loses access. Connect again using a valid key.
  • Wrong company: Ask the AI to list your firms and companies again. Company selection is per API request.
  • API limit reached: Wait for the returned retry interval. MCP requests and the underlying API calls count toward existing API limits.
  • Approval required: Finish the requested approval in API Keys & MCP in Wesley. An AI's own confirmation dialog does not replace it.