improvement(mcp): align tool discovery timeouts and retries with MCP protocol standard

[waleedlatif1]

Summary

• Raise the tools/list idle timeout from an aggressive 10s → 30s (derived from the per-server timeout, clamped to the max execution timeout) with a 60s hard cap via maxTotalTimeout, matching the MCP TypeScript SDK's DEFAULT_REQUEST_TIMEOUT_MSEC (60s). Enable resetTimeoutOnProgress + an onprogress handler so slow-but-alive servers aren't spuriously failed with McpError -32001.
• Retry read-only tools/list on transient transport errors (timeout, connection-closed, 429/5xx, session 404/400, network) with jittered backoff. tools/call stays conservative (session-error retry only) since it isn't idempotent. The SDK doesn't retry POST requests — only SSE streams — so the app owns this.
• Wire client.onerror so out-of-band transport failures (SSE disconnects, exhausted reconnects) are observed instead of silently dropped, per the SDK's Protocol.onerror contract.
• UX: map timeouts to a clear user-facing message, and surface refresh/remove failures through the standard emcn toast instead of swallowing them.

Context

A workspace MCP server whose tools/list intermittently took >10s was hard-failing discovery (-32001) even though connect/OAuth succeeded — surfacing as a stuck "Connecting…" / "Failed to discover MCP tools" with the tool count dropping from 40 → 0. The 10s cap was ~6× tighter than the SDK default and below every documented ecosystem norm; this loosens it toward the protocol standard and adds the resilience the SDK expects apps to implement themselves.

Type of Change

• Bug fix
• Improvement

Testing

• apps/sim/lib/mcp unit tests (client/service/utils) + MCP route/component/hook tests pass (116 tests).
• Added tests: tools/list timeout retries then succeeds; timeout maps to the user-facing lastError; listTools is bounded with idle timeout + hard cap + progress reset.
• tsc --noEmit clean; biome clean.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Jul 18, 2026 12:46am

[cursor]

PR Summary

Medium Risk
Changes core MCP discovery and retry behavior used by refresh and workspace tool loading; scope is bounded to read-only tools/list and error handling, with broad test coverage.

Overview
MCP tool discovery is loosened and hardened so slow-but-healthy servers stop failing spuriously and transient outages recover without user action.

tools/list now uses a 30s idle timeout (from per-server config, clamped) and a 60s hard cap, with resetTimeoutOnProgress and progress logging. Discovery retries transient failures (timeouts, connection closed, 429/5xx, stale session 404/400, network blips) with jittered backoff; persisted lastError maps timeouts to a clear user-facing message. The MCP client logs transport errors via client.onerror.

The refresh API wraps post-discovery workflow schema sync in try/catch so a sync failure still returns 200 with connected status and zero workflows updated. MCP settings shows toasts on refresh/remove failures.

^{Reviewed by Cursor Bugbot for commit 561a32f. Configure here.}

[greptile-apps]

Greptile Summary

This PR makes MCP tool discovery more resilient and improves failure reporting. The main changes are:

• Adds bounded idle and total timeouts for tools/list.
• Retries transient discovery failures with jittered backoff.
• Records transport errors reported outside normal requests.
• Maps discovery timeouts to clearer user-facing messages.
• Shows refresh and removal failures through standard toasts.
• Keeps successful refreshes from failing when workflow synchronization fails.

Confidence Score: 5/5

This looks safe to merge.

• The updated timeout calculation keeps each discovery request within the intended limit.
• The new test covers oversized server timeout settings.
• No blocking issue remains in the changed fix path.

Important Files Changed

Filename	Overview
apps/sim/lib/mcp/client.ts	Adds transport error logging and applies bounded idle and total timeouts to tool discovery.
apps/sim/lib/mcp/client.test.ts	Covers discovery timeout options and oversized configured timeout values.
apps/sim/lib/mcp/service.ts	Adds transient discovery error classification, retries, backoff, and clearer timeout messages.
apps/sim/lib/mcp/service.test.ts	Covers successful retry behavior and persisted timeout messages.
apps/sim/lib/mcp/utils.ts	Defines separate idle and total timeout constants for tool discovery.
apps/sim/app/api/mcp/servers/[id]/refresh/route.ts	Prevents secondary workflow synchronization failures from turning a successful refresh into an error response.
apps/sim/app/api/mcp/servers/[id]/refresh/route.test.ts	Covers successful refresh responses when workflow synchronization fails.
apps/sim/app/workspace/[workspaceId]/settings/components/mcp/mcp.tsx	Displays standard error toasts when server refresh or removal fails.

_{Reviews (2): Last reviewed commit: "fix(mcp): keep tools/list absolute timeo..." | Re-trigger Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 561a32f. Configure here.}