Connect an agent

Connect in three steps

1. Add the server. In an agent that supports remote HTTP MCP and browser-based OAuth, add this address:
   https://mcp.runclear.appCopy
   No API key is needed.
2. Approve access. Start the connection in your agent and sign in when the browser opens. Choose your organization, team access, and permissions. Members can choose their own teams; administrators can grant all-team access.
3. Try a read. Ask your agent to find a known contact. Then try a draft: “Create a draft proposal for this customer using our catalog. Leave it as a draft.”

You control the connection

An agent acts as you, within both its approved permissions and your current access. Each connection belongs to one organization. Contacts, companies, and product definitions are shared across that organization; deals and documents follow the selected team.

Sending documents and issuing invoices have separate permissions. Prepare the PDF and review it before sending. Granting draft access does not grant sending access.

Agents can read and add record attachments with the approved attachment permissions and access to the record. Uploading also needs HTTP/file support in your agent: files transfer outside the conversation, with a 5 MB limit and up to 10 attachments per record. Upload and download links are short-lived; keep them private.

Catalog and service gallery photos use separate gallery permissions. Agents can read, upload, reorder, edit image alt text and unlink photos for categories, product models, stock items and service offerings. Gallery uploads accept JPEG, PNG, GIF or WebP, up to 5 MB each and five images per record. Uploads verify decoded bytes, with a limit of 16 million pixels and 100 animation frames. Image alt text uses the existing label field and is limited to 50 characters through MCP; HTTP image labels retain their 280-character limit. Unlinking a gallery photo retains stored bytes to protect shared references.

Review or revoke access in Account → Agents. Administrators can review organization connections and activity in Organization Settings → Agents. Revocation blocks subsequent calls. To approve additional permissions, your client must request them during a new authorization flow; refreshing tokens or reconnecting with the previous scope list does not expand access.

Optional workflow skills

Connected agents can read these guides as MCP resources if their client supports them. You can also download a skill and install it in your harness’s skill directory, inside a folder named for the skill. They are optional; they do not add permissions.

• Client and CRM context ↓SKILL.md
• Proposals, quotes and invoices ↓SKILL.md
• Catalog setup ↓SKILL.md
• Tasks and fulfillment ↓SKILL.md
• Inventory attributes ↓SKILL.md

If something doesn’t work

New permissions missing after reconnecting? Check that your agent refreshes the server’s OAuth discovery information and requests the new scopes. Attachment tools require attachments:read or attachments:write, shown under CRM → Attachments. Gallery tools require gallery:read or gallery:write, under CRM → Gallery images, plus the parent record’s read scope and current edit permission for writes. If your client keeps an old scope list, update its requested scopes or remove and add the server again. In ChatGPT, try refreshing the connection metadata first. If new permissions are still absent, delete the custom plugin itself and re-add it; reconnecting OAuth alone may retain the old scope list. Confirm the permissions appear before approving, then refresh the client’s tool list.

No browser sign-in? Check that your client supports MCP OAuth. A tool is missing or refused? Check the connection’s permissions, your team access, and the organization’s enabled features. Repeated sign-in prompts? Check that your client saves and refreshes OAuth credentials; revoking or reloading a connection may require signing in again.