Before You Troubleshoot
Start with these basics:
The Octoparse MCP server URL is correct
Your AI client actually supports the MCP flow you are trying to use
Your account or workspace plan allows connectors/apps / MCP servers
Authentication completed successfully
The connector or server is enabled in the current environment
The requested Octoparse action is supported by your plan and task type
A lot of “MCP not working” cases are not server failures. They are usually caused by one of four things: client setup, workspace permissions, OAuth/auth, or tool visibility in the current conversation/session. That pattern is consistent with the way Anthropic, OpenAI, Microsoft, and Gemini CLI document their MCP or connector flows.
1. I can’t find the MCP or connector setting
This usually means the feature is controlled by your plan, workspace, or product surface.
For Claude, custom connectors using remote MCP are available through the connectors flow. Anthropic documents these under the connectors experience for Claude and Claude Desktop. (support.anthropic.com)
For ChatGPT, MCP-powered apps are tied to developer mode and workspace controls. OpenAI documents developer mode for building and testing MCP-powered apps under workspace-managed settings. (help.openai.com)
For VS Code, MCP support is managed under GitHub Copilot MCP server settings rather than a general connector marketplace. (code.visualstudio.com)
What to do
Confirm you are on a supported plan
Check whether an admin or owner must enable the feature first
Make sure you are looking in the right product area for that client
2. The connector or MCP server won’t save
If setup fails, the problem is usually one of these:
wrong URL or config format
unsupported auth flow
workspace policy blocking setup
platform-side timeout during initialization
For Claude, remote servers are added through Settings > Connectors. Anthropic notes that remote server setup should follow the connectors flow rather than older local config patterns. (support.anthropic.com)
For VS Code, MCP config is stored in mcp.json, either in the workspace or user profile. A malformed or misplaced config may not load. (code.visualstudio.com)
For Gemini CLI, MCP servers are defined under mcpServers, so incorrect JSON or the wrong config scope can prevent loading. (github.com)
What to do
Recheck the exact server URL or config syntax
Re-add the connector or server from scratch
Make sure you are following the client’s current setup flow
Restart the client or session after config changes
3. OAuth or sign-in fails
If you sign in but the integration still does not work, the most common reasons are:
the wrong account was authorized
the OAuth flow did not complete properly
the client did not store the granted connection correctly
the MCP server requires auth settings that were not configured
Anthropic documents OAuth support for remote MCP custom connectors, and MCP’s own authorization guidance recommends OAuth for restricted HTTP-based servers. (support.anthropic.com)
What to do
Disconnect and reconnect the integration
Re-run the auth flow in the same browser session
Confirm you authorized the correct Octoparse account
Retry after signing out of conflicting sessions
About OAuth
OAuth means the AI client does not need your Octoparse password directly. Access is granted through authorization tokens and approved permissions. (support.anthropic.com)
4. The connector is added, but the tools don’t show up
This is one of the most common issues.
Usually the integration exists, but it is not enabled in the current chat, session, or tool-enabled surface.
For Claude, connectors can exist in your account but still not be active in the current conversation. Anthropic documents enabling connectors from the Claude UI after adding them. (support.anthropic.com)
For VS Code, the MCP server may be loaded but not exposed where expected. Microsoft’s MCP docs tie usage to Copilot workflows, including agent mode and other tool-capable surfaces. (code.visualstudio.com)
For Gemini CLI, this often means the session has not picked up config changes yet. (github.com)
What to do
Check whether the connector is enabled in the current conversation
Make sure you are using the correct mode, such as agent mode or a tool-enabled chat surface
Restart the client or CLI session after config changes
Re-authorize if the tool list appears empty after auth
5. It works for me but not for teammates
This usually points to workspace permissions, not an Octoparse issue.
For Claude Team / Enterprise, Anthropic says owners may need to add the connector first, and then each user connects and authorizes it individually. (support.anthropic.com)
For ChatGPT, admin settings can control developer mode and app permissions. One user may be able to test an MCP-powered app while another user in the same workspace cannot. (help.openai.com)
For VS Code, organization-level settings can also restrict which MCP servers can be used. (code.visualstudio.com)
What to do
Check whether an owner or admin must approve the integration first
Confirm each user has completed their own authorization flow where required
Review organization-level MCP or tool access policies
6. Octoparse is connected, but actions fail
If the tool is visible but specific actions fail, the issue is often not MCP itself.
Typical causes:
the template is not supported for the requested action
the task is local-only
the export format is not available
the action requires plan features or credits you do not have
the request is trying to do too much at once
What to do
Test a small action first, such as listing templates or checking account info
Then try task creation
Then task start or stop
Then export
This helps identify whether the issue is caused by auth, permissions, plan limitations, or timeouts.
7. The request takes too long and then fails
This is usually a timeout issue.
Long prompts such as “find a template, create a task, run it, wait, then export CSV” are more likely to fail than short step-by-step prompts.
Better approach
Use a sequence like this:
Find a template
Create a task
Start the task
Check the status
Export the data
This is more reliable across almost all clients.
8. It works on desktop but not on mobile
This is often expected behavior.
Anthropic says remote MCP servers are supported on Claude and Claude Desktop, and also on iOS and Android once the remote server has already been added through claude.ai. Mobile use may differ from initial setup. (support.anthropic.com)
What to do
Add and configure the connector on web or desktop first
Then test it on mobile
Do not assume every mobile surface supports full MCP setup flows
9. You are mixing up local MCP and remote MCP
This causes a lot of confusion.
For Claude, Anthropic documents local MCP servers on Claude Desktop as a separate workflow from remote MCP connectors. They are not configured the same way. (support.anthropic.com)
For Gemini CLI and VS Code, both local and remote patterns may exist depending on how the server is hosted and defined. (code.visualstudio.com)
What to do
Confirm whether Octoparse is being connected as a hosted remote MCP server
Do not follow local server installation steps unless you are intentionally running a local server yourself
10. You are concerned about security
That concern is valid.
MCP can expose tools, resources, prompts, and actions to an AI client. VS Code explicitly notes that MCP servers can provide tools, resources, prompts, and interactive apps. (code.visualstudio.com)
What to do
Only connect to servers you trust
Review requested permissions carefully
Use the least access necessary
Disconnect and revoke access when no longer needed
Security reminder
Only connect Octoparse or any other MCP server to AI clients you trust and control. Remote MCP servers can expose powerful actions depending on granted permissions.
Client-specific troubleshooting notes
ChatGPT
ChatGPT’s MCP flow is more workspace- and admin-sensitive than most others. If the server is fine but the app is not visible or usable, check:
whether developer mode is enabled where required
whether the app is available in the workspace
whether the user is allowed to access it
whether the app is enabled in the current chat context
Claude
Claude’s model is simpler, but users often confuse local Claude Desktop MCP setup with remote custom connectors. Check:
whether the connector was added under Connectors
whether it is enabled in the current conversation
whether an owner needed to add it first
whether you are following the correct remote setup path
VS Code
Most VS Code issues come from config location, workspace trust, org policy, or using the wrong Copilot surface. Check:
whether the server loaded from the correct
mcp.jsonwhether workspace trust is blocking behavior
whether org policy restricts MCP usage
whether you are using a tool-enabled Copilot surface such as agent mode
Cursor
Cursor supports MCP, but it can have more version-specific or feature-specific quirks. Check:
whether the server is connected but not exposed in chat
whether the issue started after an update
whether restarting Cursor and reloading MCP settings helps
Gemini CLI
Gemini CLI is the most configuration-driven. Check:
whether your
mcpServersJSON is validwhether you edited the correct config scope
whether auth completed successfully
whether you restarted the CLI after the change
Quick troubleshooting checklist
If Octoparse MCP is not working, check these in order:
Correct URL or config
Supported plan or workspace access
OAuth or auth completed successfully
Connector or server enabled in the current chat or session
Correct client mode or tool-enabled surface
Organization or admin permissions not blocking use
Requested Octoparse action supported by your plan
Prompt not combining too many long-running steps
Best practice for more reliable usage
Use short, single-purpose prompts such as:
Find an Octoparse template for Amazon product listings
Create a task from this template
Start the task in the cloud
Check the latest task status
Export the latest results as CSV
This reduces timeout risk and makes troubleshooting easier.
FAQ
Why does Octoparse MCP work in Claude but not in ChatGPT?
Because the two products use different setup and permission models. ChatGPT is often more dependent on workspace admin settings and developer mode, while Claude uses a connector-based flow. (help.openai.com)
Why do tools appear in VS Code but not in a web chat client?
VS Code ties MCP more directly to Copilot workflows and tool-enabled surfaces such as agent mode. A connector can be configured correctly but still not appear where you expect in other clients. (code.visualstudio.com)
Do I need to authorize Octoparse separately on each platform?
Usually yes. Even if the same Octoparse account is used, each client may require its own connector setup and OAuth authorization session.
Does OAuth mean the AI client can see my Octoparse password?
No. OAuth is designed so the client receives authorized access without directly handling your password. (support.anthropic.com)
Why do long prompts fail more often?
Because multi-step requests are more likely to hit timeout or tool execution limits. Breaking the workflow into smaller actions is usually more reliable.