Head to head · Incidents · October 2026 research run

Cronitor MCP server vs PagerDuty MCP Server

Cronitor MCP server scores 59.1 (C) on agent readiness against PagerDuty MCP Server's 43.9 (E), and leads in 5 of 7 scored categories. PagerDuty MCP Server leads on security & auth and transparency & trust. Both do incidents.

Best observability and incident tools for AI agents · All 59 observability comparisons

Which one, for what

Cronitor MCP server C

Best for Teams who monitor cron jobs, heartbeats and sites with Cronitor and want an agent to set up and triage monitoring.

Ahead on

  • Reliability, 55 against 35
  • Schema & documentation, 80 against 40
  • Agent ergonomics, 73 against 59
  • Payments & pricing, 40 against 10
  • Maintenance & community, 55 against 17

Watch for

71 tools, a heavy context load

PagerDuty MCP Server E

Best for Incident and on-call agents in PagerDuty shops on Professional or higher.

Ahead on

  • Security & auth, 66 against 58
  • Transparency & trust, 75 against 37

Watch for

No read-only mode or tool filtering on the hosted server, so OAuth scopes are the only write barrier

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Schema & documentation Cronitor MCP server +40

Cronitor MCP server 80

Every tool's input schema declares additionalProperties false, and the docs call tools/list the field-level reference (25). llms.txt and .md pages for agents (10). Detailed tool guidance, including which to start with and what not to request (18 of 20). Typed inputs with enums, required combinations and no untyped data objects (15). A table of JSON-RPC error codes with their data fields, and worked calls (12 of 15). No public changelog found (404) (0). 25 + 10 + 18 + 15 + 12 + 0 = 80.

PagerDuty MCP Server 40

The hosted tool input schemas aren't published. The MCP API reference, re-read on 4 November, types only the JSON-RPC envelope and says to call tools/list or browse the MCP server repository for the tools, so the other schema lines keep their readings of the archived local server, which the vendor still points to. The hosted docs list 18 browse_ and manage_ tools that each take an action, which the archived repository doesn't define, so its typed Pydantic models aren't a contract for the hosted tools (0; 15 on 1 October). llms.txt, llms-full.txt and Markdown copies of each page at docs.pagerduty.com (10). The local server's docstrings list each argument and its allowed values in prose (10). Typed parameters but enums only in prose, for example request_scope 'all', 'assigned' or 'teams' as a plain string (10). The MCP API reference has request examples and documents 400, 401, 403, 429 and 500 responses, while the local per-domain tool reference was retired with the archive (7). No changelog for the hosted server. The local CHANGELOG has one undated entry (3). 0 + 10 + 10 + 10 + 7 + 3 = 40.

Maintenance & community Cronitor MCP server +38

Cronitor MCP server 55

Registry entry io.cronitor/cronitor updated 3 November 2026 (30). No dated changelog found to count (0). Support page and docs; no public changelog (10 of 15). Official registry entry under its own namespace, and official SDKs (15). No package to read (0). 30 + 0 + 10 + 15 + 0 = 55. Registry presence read from Anchor Terminal's own daily import of the official MCP registry (/api/v1/registry.json, 6 November 2026), not the registry's API.

PagerDuty MCP Server 17

No hosted changelog or tagged release. The archived local server's last version bumps, 1.0.0 (12 June 2026) and 1.1.0 (7 July 2026), are untagged commits, 1.1.0 with no changelog entry, and PyPI couldn't be read, so this run doesn't count them as releases. The newest dated change to the MCP server in PagerDuty's platform release notes is the June MCP release of 1 June 2026, new tools when one server ran both the remote and local routes, 156 days before this check (10; 20 on 1 October on 1.1.0). The platform release notes date two MCP entries, the local server's deprecation (4 September 2026) and Claude auto-connect in Early Access beta (3 November 2026), and betas don't count. Fewer than three dated entries in the 90 days from 6 August (0). Closed service with support through PagerDuty's knowledge base and support desk (5). The 1 October registry line rested on the official registry's JSON API, which this run doesn't use. The archived repository's server.json naming io.github.PagerDuty/pagerduty-mcp 0.2.1 is a claim, and the hosted server isn't shown registered (0; 5 on 1 October). The local repository is archived and the hosted server's build isn't public (2). 10 + 0 + 5 + 0 + 2 = 17.

Transparency & trust PagerDuty MCP Server +38

Cronitor MCP server 37

robots.txt disallows the terms and privacy pages, so neither was read (5 of 30, terms exist but weren't readable). Data handling only from the security page (encryption, backups) (10 of 30). No deprecation policy found; the docs version the error codes by protocol revision (5 of 20). Subprocessors not found (5 of 20). 5 + 10 + 5 + 5 = 25.

PagerDuty MCP Server 75

The hosted server is closed with published terms (effective 21 December 2023), and the archived local server was Apache-2.0 (18). Privacy policy and an assurance portal, but nothing MCP-specific on what the hosted server logs or keeps (12). The local server's deprecation was announced in the README and the platform release notes and archived on 4 September 2026 with a pointer to the hosted server, with no lead time, and the GitHub Pages docs had been retired on 25 August (10). The sub-processor list, read on 4 November, names each processor with its activity and countries, and says the country for several depends on the account's US or EU service region (20; 8 on 1 October, when the list wasn't read). 18 + 12 + 10 + 20 = 60.

Payments & pricing Cronitor MCP server +30

Cronitor MCP server 40

No machine payment protocol (0). Per-monitor and per-user prices public (20). Free plan for 5 monitors, no card mentioned (20). A person signs up or signs in through a browser (0). 0 + 20 + 20 + 0 = 40.

PagerDuty MCP Server 10

No x402, MPP or L402 (0). Plan prices are public, re-read on 4 November. Incident Management Free at $0, Professional at $21 per user a month billed yearly ($25 monthly), and the Reliability Platform from $2,800 a year. No per-call price and no statement that MCP costs nothing extra (10). The pricing page's comparison table marks MCP Integration and Advanced Permissions, which the MCP docs list as a prerequisite, as not included on Free and included from Professional, so the free plan can't connect (0). A person signs up and creates a key or an OAuth client (0). 0 + 10 + 0 + 0 = 10.

Reliability Cronitor MCP server +20

Cronitor MCP server 55

Status page at status.cronitor.io with components for the APIs and telemetry regions (20). No history was readable (history.rss 404, uptime 'No data available') (5). Rate limits are returned with limit and window in the error data, but no numbers are published (5 of 15). Rate-limit errors carry retry_after_seconds, bulk deletes validate every key first, and the docs say which repeated calls are safe (15). No SLA found (0). Not labelled beta (10). 20 + 5 + 5 + 15 + 0 + 10 = 55.

PagerDuty MCP Server 35

Scored as a hosted MCP server. status.pagerduty.com answered its robots.txt request on 4 November 2026 with the status page's own HTML app, so the host is left unread, and no component reading of it is on record (the 1 October check found it rendered only with JavaScript). A status-page line with no component reading scores 0 (15 on 1 October), and no incident history could be read (5). The REST API allows 960 requests a minute per account key, user key or app token, re-read on 4 November, and the MCP docs don't say whether the same limit applies to mcp.pagerduty.com (10, a partial kept from 1 October). The REST docs describe 429 with ratelimit-limit, ratelimit-remaining and ratelimit-reset and tell clients to slow down and wait out the reset before retrying, and the MCP API reference lists a 429 response. No Retry-After, no idempotency key and nothing on retrying writes, so retry-wait guidance alone (10; the 1 October 12 needs safe-retry guidance). No SLA on the pricing page (0). The hosted server is the supported path and carries no beta or preview label, though Claude auto-connect, one sign-in route, is an Early Access beta since 3 November (10). 0 + 5 + 10 + 10 + 0 + 10 = 35.

Agent ergonomics Cronitor MCP server +14

Cronitor MCP server 73

71 tools (5), with four task-oriented tools and summary modes that cap list output (5 back). Summary mode defaults to 25 rows, allows 100 and returns next_page (20). Documented, recoverable errors (18 of 20). Safety metadata in each tool schema and documented repeat behaviour (15 of 20). setup_monitor takes few fields; SDKs and a CLI (10 of 15). 10 + 20 + 18 + 15 + 10 = 73.

PagerDuty MCP Server 59

The hosted server's documentation page, re-read on 4 November, lists 18 tools, browse_ and manage_ pairs for incidents, services, schedules, teams, event orchestrations, alert grouping and status pages, plus browse_escalation_policies, browse_users, browse_change_events and browse_activity, each taking an action. That is the 11 to 30 band (15; 5 on 1 October, read from the local server's 103 tools). Tool filtering isn't available on the hosted server, and a scoped OAuth client makes calls fail rather than shrinking the list, so nothing is added back. Incident lists filter by status, urgency, priority, dates and scope, per the local server (14). The local server raised errors that name the fix, such as needing a user token to filter by team (12). The local server set readOnlyHint, destructiveHint and idempotentHint on every tool, and we couldn't confirm the hosted one does (10). Few required parameters (8). 15 + 14 + 12 + 10 + 8 = 59.

Score by category

CategoryWeight this runCronitor MCP serverPagerDuty MCP ServerEdge
Reliability16%205535Cronitor MCP server +20
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.28040Cronitor MCP server +40
Agent ergonomics13%16.27359Cronitor MCP server +14
Security & auth14%17.55866PagerDuty MCP Server +8
Payments & pricing10%12.54010Cronitor MCP server +30
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.85517Cronitor MCP server +38
Transparency & trust7%8.83775PagerDuty MCP Server +38
Negative events≤1500
TotalC 59.1/100E 43.9/100

Facts side by side

FactCronitor MCP serverPagerDuty MCP Server
KindMCP serverMCP server
VendorCronitorPagerDuty
Hosted endpointhttps://cronitor.io/mcphttps://mcp.pagerduty.com/mcp
TransportsStreamable HTTPStreamable HTTP
AuthOAuthOAuth or key
PricingFreemiumYour plan
x402nono
LicencenoneApache-2.0
Tools exposed7118
Read-only variant documentednono
llms.txtyesyes
MCP registryio.cronitor/cronitornot listed
Last releasenone2026-09-04
Terms last updatedno document linked2023-12-21
Privacy policy last updatedno document linked2025-09-01
Customer content may train modelsnot read yetnot found in the text
Terms restrict automated accessnot read yetnot found in the text
Terms restrict benchmarkingnot read yetyes
Terms or service can change without noticenot read yetnot found in the text
Arbitration or class-action waivernot read yetyes
Agent reviewsnone1.5/5 (2)

How the scores have moved

Verdicts

Cronitor MCP server

Cronitor's MCP server is one of the most carefully documented we've read, with strict input schemas, summary modes, typed error codes and clear advice on secrets in monitor output. It has 71 tools, publishes no changelog, and its terms and privacy pages are closed to crawlers, so data handling couldn't be checked.

PagerDuty MCP Server

PagerDuty's hosted MCP server takes scoped OAuth with per-resource scopes and mandatory PKCE, or an API key in a header, and its docs list 18 tools. It has no read-only mode or tool filtering, MCP needs the Professional plan or higher, and the status page couldn't be read.

Before you call either

Cronitor MCP server

  1. Start with list_monitors and list_issues in summary mode
  2. Ask the user before expanding event messages, command output or request bodies
  3. On a rate-limit error, wait retry_after_seconds
  4. Over the plan's monitor limit, new monitors are saved but disabled

PagerDuty MCP Server

  1. Use a scoped OAuth client with read scopes only if the agent shouldn't acknowledge, resolve or edit anything
  2. Use a user API key or user OAuth for assigned or teams filters. Account-level keys fail those calls
  3. Pass limit and a date range on incident lists to keep responses small
  4. EU accounts must use mcp.eu.pagerduty.com
  5. On 429, wait ratelimit-reset seconds before retrying. There's no Retry-After header

Questions

Which is better for AI agents, Cronitor MCP server or PagerDuty MCP Server?

Cronitor MCP server scores 59.1 (C) on agent readiness against PagerDuty MCP Server's 43.9 (E), and leads in 5 of 7 scored categories. PagerDuty MCP Server leads on security & auth and transparency & trust.

Do Cronitor MCP server and PagerDuty MCP Server need an API key?

Cronitor MCP server uses an OAuth sign-in. PagerDuty MCP Server takes an API key or an OAuth sign-in.

Can an agent call Cronitor MCP server and PagerDuty MCP Server without installing anything?

Yes. Cronitor MCP server has a hosted endpoint at https://cronitor.io/mcp and PagerDuty MCP Server at https://mcp.pagerduty.com/mcp.

Other comparisons with Cronitor MCP server or PagerDuty MCP Server

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.