Automations HTTP MCP tool calls fail with `invalid_argument`

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Starting around 2026-08-07 (JST morning) and still ongoing, Cursor Automations on our Team plan fail HTTP MCP tools/call with:

HTTP MCP tool execution failed: [invalid_argument] Error

This is not “MCP missing from the automation.” Required MCP servers are explicitly added under each automation’s Tools. The same generic error occurs across multiple MCP servers, and in one run even the built-in cursor-cloud MCP failed with the same message.

Almost all of our internal Automations are currently non-functional. Many business workflows that depend on them are blocked. We would appreciate a fix as soon as possible.

status.cursor.com showed All Systems Operational at the time of writing.

Environment

  • Product: Cursor Automations → Cloud Agents
  • Plan: Team
  • Automation scope: AUTOMATION_SCOPE_TEAM_EDITABLE (runs with a team service account)
  • First observed: 2026-08-07 ~00:47 UTC (09:47 JST); still reproducing as of this report
  • Models observed: composer-2.5, composer-2.5-fast

Observed behavior

  • HTTP MCP calls fail with the same invalid_argument error across multiple configured MCP servers (e.g. Slack, Linear, Notion, Datadog, Sentry — depending on the automation)
  • Built-in cursor-cloud MCP also failed with the same error on one of the runs below
  • On one run, the agent reported that GetMcpTools showed servers as ready, but the tools array was empty
  • Repo checkout / code exploration still worked; only HTTP MCP execution failed, so external write/update steps were skipped

Steps to Reproduce

  1. Ensure Team MCP servers are enabled in the dashboard, and explicitly add the required MCP servers under an Automation’s Tools.
  2. Trigger the Automation (e.g. via Slack).
  3. Open the resulting Cloud Agent run.
  4. Observe HTTP MCP tool calls fail immediately with:
HTTP MCP tool execution failed: [invalid_argument] Error

This reproduces across multiple Automations that depend on HTTP MCP.

Expected Behavior

HTTP MCP tools enabled on an automation should execute successfully, or return a clear, actionable error. They should not fail with a generic [invalid_argument] Error with no detail.

Operating System

MacOS

Version Information

N/A — issue is in Cursor Automations / Cloud Agents (web), not a local IDE/CLI/iOS client build.
Observed Cloud Agent models: composer-2.5, composer-2.5-fast

For AI issues: which model did you use?

composer-2.5, composer-2.5-fast

For AI issues: add Request ID with privacy disabled

bc-27e01687-9d1b-59d6-9872-3576e2b8e9aa
bc-8eee466e-b2d1-5ec0-a2c1-115f71b98f1c
bc-aa2f8d8a-b964-5965-aabc-21b6ab37267f

Additional Information

What we already ruled out

  1. MCP not added to the automation tool list (they are added)
  2. Local .cursor/mcp.json dependency (using dashboard Team MCP / automation MCP actions)
  3. Platform-wide status incident (status page green)
  4. Single MCP provider outage (multiple MCP servers + built-in cursor-cloud)

Impact

Almost all of our team’s Automations that rely on HTTP MCP are broken right now. This blocks multiple production business flows. Please prioritize a fix.

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

7

Hey there,

You’re not doing anything wrong. This isn’t coming from your MCP configuration (which is why the same servers work fine in your local coding agents) - it’s an issue on our side affecting HTTP MCP tool calls in Automations. We can see the failures on the runs you shared, we’ve let the team know, and this is an issue we’re tracking.

No action needed on your end. I’ll post here as soon as I have an update - and if there’s a particular workflow that’s especially time-critical, let me know and I’ll flag it.

9

Hey! This should be sorted out now - the invalid_argument failures on HTTP MCP tool calls in automations were caused by a change on our side that’s since been rolled back. Give it another run and let me know if you still see any tool calls failing. Thanks for the clear report and the patience while we got it back to normal.

10

Thanks Mohit — confirmed, HTTP MCP tool calls in Automations are working again on our side.