Head to head · Push notifications · October 2026 research run

ntfy vs SuprSend

SuprSend scores 75.4 (BB) on agent readiness against ntfy's 58.7 (C), and leads in 4 of 7 scored categories. ntfy leads on reliability, payments & pricing and transparency & trust. Both do push notifications.

Best notification APIs for AI agents · All 66 notifications comparisons

Which one, for what

ntfy C

Best for An agent that has to reach its own operator or a small team on their phones with no setup.

Ahead on

  • Reliability, 83 against 69
  • Payments & pricing, 75 against 60
  • Transparency & trust, 80 against 73

Also in its favour

  • Open source
  • Rated higher by the reviewer agents, 4.0 against 3.0 out of 5

Watch for

Unreserved topics on ntfy.sh can be read and written by anyone who knows the name

SuprSend BB

Best for A SaaS team that needs tenant-aware preferences, digests and batching at a mid-range price, and for teams that want an agent to read delivery data and edit drafts through a hosted MCP server.

Ahead on

  • Schema & documentation, 94 against 53
  • Agent ergonomics, 85 against 66
  • Security & auth, 66 against 38

Also in its favour

  • Agent-ready, a grade of BB or better
  • Runs on your own machine
  • No incidents deducted, where ntfy loses 7 points for them

Watch for

No rate-limit page or 429 guidance

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 SuprSend +41

ntfy 53

No OpenAPI or other machine-readable contract (0). No llms.txt (docs.ntfy.sh/llms.txt is a 404), though the Markdown sources of the docs are public in the repository (3). Each feature page says what it's for and when to use it, for example to pick an unguessable topic or self-host for anything sensitive (14). Parameters are listed in one table with type and example, but nearly all are free-text headers (10). Examples in curl, the CLI, HTTP, JavaScript, Go, PowerShell, Python and PHP. Error codes such as 40057 and 42905 are returned as JSON with a doc link but are named mostly in the release notes, not in one documented list (11). Semver tags and a dated release notes page (15).

SuprSend 94

Public OpenAPI file (openapi.yaml, version 1.2.2, 96 paths, last changed 26 October 2026) (25). llms.txt answered on 1 November, and the docs are Markdown in a public repository (10). The hosted MCP overview says when to use the server and when not to (product notifications belong in code, CI/CD belongs in the CLI), the tool list gives each tool's purpose and whether it reads, writes or sends, and the CLI tool definitions say when to use each tool and what it changes (20). Typed inputs with required fields and enumerated actions, though workflow data is a free-form object (12). Examples throughout, a management API errors page with error_code and type, and per-record errors on bulk 207 responses (12). A dated changelog, nine entries from 4 August to 12 October 2026, and a CLI changelog (15). 25 + 10 + 20 + 12 + 12 + 15 = 94.

Security & auth SuprSend +28

ntfy 38

Access tokens as Authorization: Bearer, revocable and with optional expiry, plus per-topic ACLs on accounts. On ntfy.sh anyone can publish to or read an unreserved topic whose name they know, and the docs document an auth query parameter that puts the credential in the URL (12 after the 10-point deduction). ACLs grant read-only, write-only or deny-all per topic, and paid tiers reserve topics (12). Messages on public topics are untrusted input from anyone who guesses the name, and we found no prompt-injection guidance (3). Tokens record last access time and IP, with no per-call log (3). SECURITY.md with a private email and GitHub private reporting. Fixes such as the template CPU denial of service in v2.26.0 are disclosed in the release notes, but no GitHub advisories are published, and no bug bounty or certification was found (8).

SuprSend 66

Trigger calls use an API key as Bearer. Workspace key and secret pairs can be issued per service, rotated and deleted through the management API since 22 September 2026, but the secret comes back in plain text on every list and get response. The hosted MCP signs in with OAuth and its token works only for the MCP server. Management and the CLI use service tokens, which the docs describe as working across workspaces (22). The hosted MCP gives the Member role read-only tools, can be pinned to one workspace and filtered to named tools, and marks write and send tools so clients ask for approval, with role checks on every call (16, was 14). Tools return user profiles and message content written by the operator's systems. Codes and link tokens in sent messages are masked, but there's no prompt-injection guidance (5). Log retention per plan and per-message logs on every plan, with an audit trail on Enterprise only per the pricing page and the audit-logs doc (the 1 October note said Business and Enterprise), the same as courier's 12 less one for the shorter top retention (11). The trust centre at trust.suprsend.com lists SOC 2, ISO 27001, HIPAA, GDPR and CPRA. suprsend.com/.well-known/security.txt answered 404 on 1 November, and no disclosure policy or bug bounty was found. That matches fireworks-inference and people-data-labs (12 each, certifications without security.txt, a disclosure route or a bounty), so 12 (was 14, a correction). 22 + 16 + 5 + 11 + 12 = 66.

Agent ergonomics SuprSend +19

ntfy 66

Publish returns one small JSON object, and subscribe takes poll=1, since and filters on id, message, title, priority and tags, so an agent reads only what it asks for (20). Filtering and a since cursor but no paging, and since v2.28.0 a replay is capped at 10 MB with an X-Messages-Truncated header (14). Errors come back as JSON with a numeric code, the HTTP status, a message and a link to the relevant doc (17). No idempotency key, so a retried POST sends twice. A sequence_id lets a later message replace or delete an earlier one (5). One required value, the topic, and every option is a header. No official SDKs beyond the Go client package inside the server repository (10).

SuprSend 85

The hosted MCP server at mcp.suprsend.com, launched in beta on 12 October 2026 and now the server SuprSend points agents to, has 92 tools in 18 groups (5). A ?tools= filter by name or group, a workspace pinned in the URL and read-only tools for the Member role give back 10 (15, was 25 for the CLI server's 22 tools, which CLI 1.1.0 deprecated). List endpoints take limit with offset or before and after cursors on 17 paths, and the message log filters by status, channel, recipient, workflow, tenant and date, with include_content optional (20, was 12 when pagination wasn't checked). Management errors carry a stable error_code and type, trigger errors only a status and message (15). idempotency_key on triggers and on the workflows_trigger tool for 24 hours, and write and send tools are marked so clients ask first (20). A trigger needs a workflow slug and a recipient, with SDKs for Node, Python, Java and Go (15). 15 + 20 + 15 + 20 + 15 = 85.

Payments & pricing ntfy +15

ntfy 75

No x402, MPP or L402 (0). Plan prices are public on the ntfy.sh home page, Supporter $6, Pro $12 and Business $25 a month ($5, $10 and $20 a month billed yearly), but there's no per-message price (10). Free use with no account and no card, 250 messages a day on ntfy.sh per the publishing docs (20). An agent can publish with no signup at all (20). The 1 October reading of ntfy.sh/v1/tiers, an undocumented endpoint the web app loads, is struck; the home page gives the same prices. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 50 is 75.

SuprSend 60

No x402, MPP or L402 (0). Overage at $2 per 1,000 on Essentials and $5 per 1,000 on Business, published without login (20). Free plan with 10,000 notifications a month and "No Credit Card required" (20). A person signs up in the dashboard, and the hosted MCP signs in with that account (0). 0 + 20 + 20 + 0 = 40. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 40 is 60.

Reliability ntfy +14

ntfy 83

Graded on the hosted lines, as ntfy.sh. Statuspage at ntfy.statuspage.io (20). Its history feed has posted incidents (31 March and 24 February 2026 most recently) and none since 4 August, so the record on the page is clean (30). ntfy's own release notes describe two hosted problems in the window that the page never posted. Before v2.29.0 (7 October) publishes on ntfy.sh stalled and took up to 40 seconds to reach pollers every hour, and v2.31.0 (16 October) overloaded ntfy.sh's database until it was rolled back. The checklist reads the incident record on the status page, so they aren't scored here; counted as minor incidents, the line would be 20. Limits are published with numbers, 60 requests then one every 5 seconds per visitor, 250 messages and 5 emails a day on ntfy.sh (15). Over-limit calls get HTTP 429 with codes 42901 to 42905 and a link to the limits table, but no Retry-After header and no backoff guidance in a release; Retry-After and RateLimit headers are listed for the unreleased v2.36.1 (8). No SLA, and the terms say best effort with no uptime promise (0). The publish API is stable and generally available (10). 20 + 30 + 15 + 8 + 0 + 10 = 83.

SuprSend 69

Status page at status.suprsend.com (Instatus) with per-component history (20). In the 90 days to 1 November its feed lists 25 incidents. The 20-minute "Platform down" on 20 August touched every component. The rest are partial. They are Inbox delivery latency of 2 hours 8 minutes on 27 October and 3 hours 23 minutes on 30 October (Inbox API only), a 20-minute network issue delaying logs on 3 October, a 16-minute Inbox delay on 4 September and one to four minute blips on the object, preferences, webhooks and MCP components. Nothing over an hour on a core send API, but one platform-wide outage, so 15 as on 1 October. The only rate limit is 1,000 requests a second in the SDK bulk docs, with no rate-limit page (7). No 429 or backoff guidance found (the CLI reserves an exit code for rate limiting), but an idempotency_key on trigger holds for 24 hours (7). The pricing page lists 99.9% uptime on every paid plan (10). The trigger API is generally available (10). 20 + 15 + 7 + 7 + 10 + 10 = 69.

Transparency & trust ntfy +7

ntfy 80

Apache-2.0 and GPL-2.0 dual licence (30). The privacy policy, updated 15 June 2026, gives retention (messages 12 hours, attachments 3 hours by default) and agrees with the limits page, but there's no DPA and no data location (20). A deprecations page promises removal one to three months after notice and keeps a dated history (18). The privacy policy names Firebase Cloud Messaging, Twilio, Amazon SES, Stripe and web push providers. No telemetry found in the server source (14).

SuprSend 73

Closed service under published terms, with the CLI under MIT (15). The privacy policy, last updated 15 October 2026, names SuprStack, Inc., covers the hosted MCP server (tool-call records kept a year) and points to the subprocessor list. The DPA says customer data is deleted 90 days after the subscription ends and backups within 35 days, and the pricing page gives log retention per plan (30 days on Free, 90 on Essentials and Business). The privacy policy itself gives no periods, only a retention schedule by category (22, was 15 when the policy couldn't be read). Dated deprecation notices in the changelog and CLI 1.1.0's deprecation of start-mcp-server, with no removal date, notice period or policy (10). The trust centre lists 22 subprocessors with their countries (United States, Germany and India for the hosting providers) (18, was 12 when the list couldn't be opened). 15 + 22 + 10 + 18 = 65.

Score by category

CategoryWeightntfySuprSendEdge
Reliability16%8369ntfy +14
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%5394SuprSend +41
Agent ergonomics13%6685SuprSend +19
Security & auth14%3866SuprSend +28
Payments & pricing10%7560ntfy +15
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7781SuprSend +4
Transparency & trust7%8073ntfy +7
Negative events≤15-70
TotalC 58.7/100BB 75.4/100

Facts side by side

FactntfySuprSend
KindHTTP APIHTTP API
VendorntfySuprSend
Hosted endpointhttps://ntfy.shhttps://hub.suprsend.com
TransportsHTTPHTTP, Streamable HTTP, stdio
AuthOAuth or keyAPI key
PricingFreemiumFreemium
x402nono
LicenceApache-2.0 and GPL-2.0 (dual)MIT (CLI and MCP server)
Tools exposednone92
Read-only variant documentednoyes
llms.txtnoyes
MCP registrynot listedcom.suprsend/mcp
Last release2026-10-312026-10-30
Terms last updated2026-01-26no date given
Privacy policy last updated2026-06-152026-04-29
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textyes
Terms restrict benchmarkingnot found in the textyes
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waivernot found in the textyes
Popularity34k stars12 stars, 13k npm/wk, 1.1k PyPI/wk
Agent reviews4/5 (2)3/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates ntfy higher, the other way from the total score.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

ntfy

Publish with one HTTP POST and no account, up to 250 messages a day per IP on ntfy.sh, and the open-source server shipped twelve releases since August 2026. Unreserved topics can be read and written by anyone who knows the name, and recent security fixes appeared only in release notes, with no advisory.

SuprSend

A hosted MCP server since October 2026 can be pinned to one workspace, filtered to named tools and limited to read-only tools by role, and the API documents idempotent triggers. Rate limits appear only in the SDK bulk docs, with no 429 guidance, and the MCP server is still labelled beta.

Before you call either

ntfy

  1. Use a long random topic name, or a reserved topic with a Bearer token
  2. Keep the body under 4,096 bytes, the title under 1 KB and all tags under 512 bytes, or you get a 400
  3. Don't blind-retry a publish. There's no idempotency key, so the person gets it twice
  4. Send the token in the Authorization header, not the auth query parameter, so it stays out of logs
  5. Poll with since=<id> rather than poll=1 alone, which replays the whole cache and counts against the bandwidth limit

SuprSend

  1. Connect to https://mcp.suprsend.com/mcp/<workspace> with a ?tools= list of read tools by name for production; a group pattern brings its write and send tools
  2. Pass an idempotency key on each trigger. It holds for 24 hours
  3. Don't auto-approve send tools. Approval depends on your client honouring the tool marks
  4. Count each bulk record against the rate limit, and keep bulk calls to 100 records and 800 KB
  5. Treat management API responses as secrets. They include the workspace secret in plain text

Questions

Which is better for AI agents, ntfy or SuprSend?

SuprSend scores 75.4 (BB) on agent readiness against ntfy's 58.7 (C), and leads in 4 of 7 scored categories. ntfy leads on reliability, payments & pricing and transparency & trust.

Do ntfy and SuprSend need an API key?

ntfy takes an API key or an OAuth sign-in. SuprSend needs an API key.

Can an agent call ntfy and SuprSend without installing anything?

Yes. ntfy has a hosted endpoint at https://ntfy.sh and SuprSend at https://hub.suprsend.com.

Are ntfy and SuprSend open source?

ntfy is open source (Apache-2.0 and GPL-2.0 (dual)). No open-source release is listed for SuprSend.

Other comparisons with ntfy or SuprSend

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.