// Query and analyze Cloudflare Workers traces through the general Cloudflare API MCP server. Use when diagnosing Workers traces, distributed tracing spans, observability events, subrequest chains, or when an agent needs Cloudflare trace data programmatically.
Create a new Cloudflare app in this repo with the expected package scripts, Doppler shape, and CI workflow wiring.
Debug MCP servers with the official MCP Inspector UI and CLI. Use when an MCP server will not connect, auth fails, tools/resources/prompts are missing, or a tool call needs to be reproduced outside Codex/Claude.
Drizzle ORM database migration workflow for packages that still use Drizzle (agents, example). Do NOT use for apps/os — that app uses sqlfu/D1.
Monitoring workflow for architect: Cloudflare Worker checks, task updates, and Slack escalation.
Debug failures in the Cloudflare OS worker using the Cloudflare MCP server. Use when diagnosing 500s, missing logs, or request-level regressions in production or preview.
PostHog alert responder workflow with required Slack deep-links and escalation rules.
| name | cloudflare-traces |
| description | Query and analyze Cloudflare Workers traces through the general Cloudflare API MCP server. Use when diagnosing Workers traces, distributed tracing spans, observability events, subrequest chains, or when an agent needs Cloudflare trace data programmatically. |
Use Cloudflare's general API MCP server only:
https://mcp.cloudflare.com/mcpcloudflare-api or cloudflareDocs: https://developers.cloudflare.com/agents/model-context-protocol/mcp-servers-for-cloudflare/
Use the general Cloudflare MCP tools:
mcp__cloudflare_api__.searchmcp__cloudflare_api__.executeexecute requires account_id when the token can see multiple accounts. Pass it
as the tool argument, not in the MCP URL.
Use these as the code strings passed to mcp__cloudflare_api__.execute.
async () => {
return cloudflare.request({
method: "POST",
path: `/accounts/${accountId}/workers/observability/telemetry/keys`,
body: {
datasets: ["otel"],
from: Date.parse("2026-05-06T15:37:00.000Z"),
to: Date.parse("2026-05-06T15:40:00.000Z"),
keyNeedle: { value: "trace", matchCase: false },
limit: 50,
},
});
};
view: "traces":async () => {
return cloudflare.request({
method: "POST",
path: `/accounts/${accountId}/workers/observability/telemetry/query`,
body: {
queryId: "trace-summary",
timeframe: {
from: Date.parse("2026-05-06T15:37:00.000Z"),
to: Date.parse("2026-05-06T15:40:00.000Z"),
},
view: "traces",
limit: 20,
parameters: {
datasets: ["otel"],
filters: [
{
key: "traceId",
operation: "eq",
type: "string",
value: "250178b64271952ffb1ed1711133cf78",
},
],
},
},
});
};
Fetch spans with view: "events" and the same exact traceId filter.
Summarize span events by source.name and $metadata.transactionName.
See EXAMPLES.md for the exact codemode block that returned
trace 250178b64271952ffb1ed1711133cf78 from os-preview-2 with 686 spans.
If this returns no rows:
$metadata.traceId as well as traceIdaccount_id matches the worker's accountview: "events" with needle: { value: traceId } as a fallbackInclude trace id, service/script name, timeframe, request URL, span counts, error spans, longest repeated chain, and whether the failure surfaced to the caller.