Find answers to common questions about the MCP protocol and MCP services.
MCP Protocol
- How is the MCP protocol implemented? MCP is an open-source standard protocol proposed by Anthropic. For detailed implementation, refer to the official MCP website.
Cloud-Deployed MCP Services
- Firecrawl/EverArt MCP Server returns an error: “Request quota exceeded” or “Insufficient balance”? Ensure you have entered a valid API key from the service provider and that your free quota or account balance is sufficient.
- Why does my Notion/GitHub MCP Server fail with a “permission denied” error? Grant sufficient permissions to your API key. Refer to the documentation for the corresponding MCP service for specific instructions.
Custom MCP Services
- Can all MCP services be deployed and used successfully? Not guaranteed. MCP service providers may modify or discontinue their services. Alibaba Cloud Bailian provides only access channels and does not guarantee long-term availability.
-
My custom MCP service failed to deploy — how do I troubleshoot?
Please verify:
- The MCP service runs correctly locally.
- The MCP service supports cloud hosting (i.e., does not require browser or local application communication).
- Installation method and configuration code are correct during deployment.
- Function Compute (FC) and related permissions are enabled, and your Alibaba Cloud primary account has no outstanding payments.
-
Can I deploy an MCP service I developed myself?
- Node.js: Publish your package to the public npm registry and deploy using
npx <package-name>. - Python: Publish your package to PyPI and deploy using
uvx <package-name>. - Already hosted on a remote server: Connect via Server-Sent Events (SSE).
- Node.js: Publish your package to the public npm registry and deploy using
- Can Alibaba Cloud Bailian MCP services be used in other MCP clients (e.g., Cline, Cherry Studio)? Yes. MCP supports both internal platform integration (e.g., within Agents or Workflows) and external integration into third-party applications (e.g., Cherry Studio, Cursor). See External Invocation for more information.
- Can Alibaba Cloud Bailian MCP services access local databases? Not supported. Currently, MCP services cannot access local resources on your machine.
- Why can’t my MCP service access remote resources (e.g., cloud databases)? MCP services are hosted on Alibaba Cloud Function Compute (FC), which does not provide a fixed public egress IP address. Configure an IP whitelist for your remote resources (e.g., cloud databases) to include FC’s IP ranges, or establish VPC peering.
- Can I deploy an MCP Server hosted in a private npm registry to Alibaba Cloud Bailian? Not supported. Please publish your package to the public npm registry, or switch to SSE-based connectivity.
- Will my custom MCP Server, once deployed, be accessible to others? No. Your custom MCP service is accessible only to your Alibaba Cloud primary account and authorized RAM users.
- Why can’t some MCP services be custom-deployed on Alibaba Cloud Bailian? Because they require access to local resources (e.g., files, hardware), which are inaccessible in the cloud environment. We recommend deploying such services locally.
-
I’ve updated my MCP Server version — will the deployed custom MCP service on Alibaba Cloud Bailian update automatically?
No. Services deployed via
npx/uvxrequire manual redeployment after version updates.
Custom MCP Service Error Codes and Troubleshooting
| Error Code | Issue & Possible Causes | Troubleshooting Steps |
|---|---|---|
| 11200044 - MCP_CONNECTION_REFUSED | Client fails to establish connection to target service; connection is refused. Possible causes: service not running; port not open or blocked by firewall; network unreachable or misconfigured address. | Run curl <service-url> to test connectivity; check downstream service logs to confirm MCP service is running; review gateway/firewall configurations for IP whitelisting or connection rejection rules. |
| 11200045 - MCP_CONNECTION_TIMEOUT | Connection establishment times out. Possible causes: network latency or packet loss; high server load or slow response; unstable routing/network; long chain when accessing through proxy. | Retry 2–3 times: if intermittent success occurs, it's likely a network issue; if consistently failing, inspect gateway/firewall configurations along the downstream service path. |
| 11200046 - MCP_REQUEST_TIMEOUT | Connection established but response timed out. Possible causes: server-side processing taking too long or resource exhaustion; insufficient bandwidth or high latency delaying response. | Retry 2–3 times. If still failing: split business logic to reduce per-request latency or make operations asynchronous; for npx/uvx deployments, enable logging in Function Compute console to inspect errors; consider switching to “Ultra-Fast Mode” to reduce cold-start latency. |
| 11200047 - MCP_NETWORK_ERROR | Network reachability, DNS resolution, or connection interruption issues. Possible causes: DNS resolution failure or misconfiguration; connection closed mid-stream by peer or intermediate device; destination host unreachable or routing anomaly. | Retry 2–3 times. If still failing: run nslookup to verify DNS resolution; change DNS servers if needed and retry; validate the url field in your config and test with curl <url>. |
| 11200048 - MCP_SSL_ERROR | TLS/SSL handshake or certificate validation failure. Possible causes: expired certificate, domain mismatch, incomplete certificate chain; client doesn’t trust server certificate (e.g., self-signed); incompatible TLS version or cipher suite; proxy/gateway rewrites certificates. | Open the service URL in a browser to check for certificate expiration or domain mismatch; configure trust properly for self-signed certificates; disable proxy and test direct connection to rule out certificate rewriting. |
| 11200049 - MCP_SERVER_HTTP_UNAUTHORIZED | HTTP 401 — request lacks valid authentication or credentials have expired. Possible causes: missing or incorrectly passed auth tokens/API keys; token expired, revoked, or malformed; required headers or cookies not provided. | Consult downstream MCP service docs for auth requirements; correctly inject auth info (e.g., Authorization header) in your MCP service config; test direct connectivity using curl. |
| 11200050 - MCP_SERVER_HTTP_FORBIDDEN | HTTP 403 — authenticated but unauthorized. Possible causes: current account/role lacks permissions for this MCP service or instance; IP/source restrictions apply. | Review downstream MCP service documentation for authorization rules; verify configuration correctness using curl. |
| 11200051 - MCP_HTTP_RATE_LIMIT | HTTP 429 — request rate exceeds limit. Possible causes: excessive requests in short time; quota exhausted for current account or instance. | Reduce call frequency; respect Retry-After header if present; contact MCP service provider to request higher quotas if needed. |
| 11200052 - MCP_HTTP_CLIENT_ERROR | Other HTTP 4xx errors (excluding 400/401/403/404/405/429). Possible causes: invalid HTTP method, malformed URL, incorrect headers/body; oversized request body or excessively long URL; server-side timeout (e.g., 408). | Inspect full response status and error message; consult downstream MCP service documentation; use curl to examine complete request/response and check downstream service logs. |
| 11200053 - MCP_HTTP_SERVER_ERROR | HTTP 5xx — internal server or gateway error. Possible causes: MCP service or gateway crash, overload, or maintenance; dependent downstream services unavailable. | Retry 2–3 times. If still failing: inspect response details and downstream MCP service logs; for npx/uvx deployments, view runtime logs in Function Compute console (log service must be enabled first); use curl to capture full request/response. |
| 11200054 - MCP_PROTOCOL_ERROR | Response cannot be parsed according to MCP protocol. Possible causes: server returns standard JSON-RPC error (e.g., parse error, invalid request, method not found, invalid params, internal error); protocol or version incompatibility. | Fix request format, method name, or parameters based on JSON-RPC error code; ensure type matches endpoint path: "sse" → /sse, "streamableHttp" → /mcp. |
| 11200055 - MCP_SESSION_NOT_FOUND | Session does not exist or has expired. Possible causes: server restart or session expiry while client uses stale session ID; session ID incorrectly passed or lost; routing policy causes requests to hit different instances. | Improve downstream MCP service adherence to standard MCP session behavior; check logs — unexpected restarts may cause session loss. |
| 11200056 - MCP_UNKNOWN_ERROR | Unclassified exception carrying raw error details. Possible causes: unhandled exceptions from runtime environment or dependencies. | Analyze raw error message and context; for npx/uvx deployments, enable logging in Function Compute console to inspect runtime logs. |
| 11200057 - MCP_INIT_TIMEOUT | No readiness signal received during initialization. Possible causes: server fails to push init data promptly; network latency or packet loss; proxy/gateway misconfigures long-lived connections. | Retry 2–3 times. If still failing: verify access URL and transport type (sse or streamableHttp) in MCP Management Console; test connectivity with curl <access-url>; configure reverse proxies to support persistent connections and set adequate timeouts; for npx/uvx, check startup logs in Function Compute console — cold starts in Basic Mode may cause delays; consider Ultra-Fast Mode. |
| 11200058 - MCP_SERVER_HTTP_METHOD_NOT_ALLOWED | HTTP 405 — endpoint rejects the used HTTP method. Possible causes: mismatch between transport type and endpoint (e.g., SSE used with Message endpoint); unsupported method applied to same URL (e.g., GET instead of POST). | Confirm type matches endpoint: "sse" → GET /sse, "streamableHttp" → POST /mcp; correct misconfigurations in MCP Management Console and redeploy; prefer standard MCP SDKs for downstream service implementation. |
| 11200059 - MCP_SERVER_HTTP_NOT_FOUND | HTTP 404 — requested path not found. Possible causes: incorrect access URL path (e.g., missing prefix); routing/gateway fails to forward path to MCP service; instance offline or improperly deployed. | Verify type ↔ endpoint alignment; check service status in MCP Management Console — redeploy if needed; test connectivity using curl. |
| 11200060 - MCP_SERVER_HTTP_BAD_REQUEST | HTTP 400 — malformed request or invalid parameters. Possible causes: request body not valid JSON or non-compliant with MCP/JSON-RPC spec; missing or incorrectly typed required parameters; headers or query params fail server-side validation. | Validate field types and required fields against service expectations; test connectivity using curl. |
Integration with Agents / Workflows
-
Why can’t my Agent application invoke MCP services, or why does it report parameter errors?
Agents determine invocation and parameters based on prompts. Try:
- Refining your prompt to clarify intent.
- If ineffective, upgrade to a stronger reasoning model (e.g., Qwen3 series).
-
Does invoking MCP increase input or output tokens for the model?
Yes. MCP invocation may increase both input and output token counts:
- Input tokens increase: Content retrieved from the MCP service becomes part of the model’s context, directly increasing input token count.
- Output tokens increase: While MCP calls themselves don’t generate output tokens, richer context may lead the model to produce longer, more detailed responses — indirectly increasing output token usage.
- Can MCP services be integrated when calling the Qwen API directly? No. Alibaba Cloud Bailian MCP services are designed to be used exclusively within Agents or Workflows, and cannot be integrated directly into standalone Qwen API calls.