TinyFish

# Fix list: TinyFish

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/tinyfish, the October 2026 research run, assessed 8 October 2026. Grade B, 67.7 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on TinyFish: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 60 out of 100, up to 8 more on the total

Why it scored 60: Graded as a hosted API. agent.tinyfish.ai/status is the vendor's own page with five components and 90 daily bars each, but no incident write-ups (15 of 20). From 11 July to 8 October 2026 the Agent API shows no downtime. The Browser API shows 420 minutes across six days, with 120 on 28 September. The Search API shows 1,500 minutes across seven days, with 420 on 7 August, and the Fetch API 240 on 27 August. Several outages of an hour or more, so 5 of 30, kept above 0 for the clean Agent API record. Limits are published (Search 30 requests a minute, Fetch 150 URLs a minute, 2 concurrent Agent runs, 5 concurrent Browser sessions) (15). Search and Fetch 429s carry `Retry-After` and `X-RateLimit-*` headers, the docs give backoff guidance and the SDKs retry 429 and 5xx. The Agent pending-run 429 has no `Retry-After` and run creation has no idempotency keys (11). The pricing page lists a 99.99 per cent uptime SLA for Enterprise, with no SLA document found (5). Browser, Search and MCP were declared generally available in June 2026 and the API is `/v1`. The SDKs are 0.x and some Agent settings are beta-gated (9).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Security & auth, 57 out of 100, up to 7.5 more on the total

Why it scored 57: REST calls take an `X-API-Key` header. Keys are created and deleted in the dashboard and shown once, with no scopes found. MCP uses OAuth 2.1, and run IDs are separated between CLI or REST tokens and MCP tokens (21). Vault credentials are opt-in per run with `use_vault` and can be limited with `credential_item_ids`, and the docs say the agent never sees passwords. No read-only key or approval step for agent actions was found, and `max_steps` is beta-gated (9). Pages are untrusted content. The plugin page mentions safety rules for untrusted content, and the API docs have no prompt-injection guidance (3). Each run keeps per-step screenshots and optional HTML snapshots, usage lists exist for Search, Fetch and Browser sessions, and audit logs are listed under Enterprise (11). security.txt is valid until 1 September 2027, and the disclosure policy v1.1 of 15 September 2026 has safe harbour and discretionary rewards with no formal bounty. The terms cite an ISO 27001 certification, and the trust centre that would show it is drawn by script and was unread (13).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Payments & pricing, 55 out of 100, up to 5.6 more on the total

Why it scored 55: `POST /v1/wallet/top-up` on agent.tinyfish.ai answers 402 with a Machine Payments Protocol challenge and takes a Stripe Shared Payment Token for $10 to $500. It refills a prepaid wallet instead of paying per call, needs an existing API key, and the docs say it is rolling out gradually, so partial credit (15). Per-unit prices are public, $0.016 an Agent step and $0.002 a Browser minute, with Search and Fetch at $0 (20). Search and Fetch are free within daily allowances and new accounts start with $8 of wallet funds, no card required per the pricing page (20). A person has to sign up in a browser to get a key or complete OAuth (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 4. Transparency & trust, 59 out of 100, up to 3.6 more on the total

Made of editorial 32, provenance 85.

Why it scored 59: The platform is closed with published terms, effective 17 August 2026. The MCP proxy and cookbook are MIT, and the SDK packages state no licence in npm or PyPI metadata (14). The terms allow Tiny Fish to review runs and to train and improve its models on customer data, which is stated plainly. The privacy policy of 14 December 2025 gives no retention periods and names TinyFish Inc. in Cupertino, while the terms name Tiny Fish, Inc. in Los Altos. No public DPA was found (10). No deprecation policy was found. The docs flag two parameters as deprecated without dates, and the terms give 30 days' notice of material changes to the terms (5). No subprocessor list or hosting location was found on readable pages. The enterprise page says data residency is agreed by contract, and the trust centre was unread (3).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: tinyfish.ai, registered 2023-02-06 (3 years) (7 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 3 clauses that cost points (3.1 of 10)

## 5. Agent ergonomics, 79 out of 100, up to 3.4 more on the total

Why it scored 79: Fetch takes an output format, include and exclude selectors, a cache age and a per-URL timeout, and run reads choose whether screenshots come back as URLs or inline. The MCP server has 28 tools per its repository, with a `guide_next_step` tool (18). Run lists take a cursor, a limit of 1 to 100 and filters for status, goal text and dates, and Search pages from 0 to 10 (18). Errors carry a code, a message, optional `details` and a `request_id`, and wallet denials add `non_retryable` and balance fields (19). The docs say run creation doesn't support idempotency keys and a retry may create duplicate runs. Cancels and session deletes are idempotent and a replayed wallet top-up is credited once. We couldn't read MCP annotations without signing in (9). A run needs only `url` and `goal`, a browser session needs nothing, and there are official Python and TypeScript SDKs and a CLI (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 6. Schema & documentation, 85 out of 100, up to 2.4 more on the total

Why it scored 85: Graded on the REST API. OpenAPI 3.0.0 documents are linked from `llms.txt` for five APIs. We read two, the main document with 28 operations and the Browser document with 3 (25). `llms.txt`, a 460 KB `llms-full.txt`, a Markdown twin of every page and a single page for coding agents (10). All 31 operations we read carry a description, several say which endpoint to pick instead, and the MCP tool definitions state when to use and when not to use each tool (17). 189 enums in the main document, ranges such as `max_steps` 1 to 500, and unknown fields rejected. 730 of 1,068 properties carry a description, `goal` is free text by design and the MCP `search` tool takes a JSON-encoded `fetch` string (12). 698 examples in the main document, every operation documents an error response, and an error-code page gives a fix for each code (14). The path is versioned `/v1` and the document is 1.0.0. The Pulse changelog stops at 22 June 2026 and the docs have no API changelog (7).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Maintenance & community, 80 out of 100, up to 1.8 more on the total

Why it scored 80: `@tiny-fish/cli` 0.48.1 was published to npm on 7 October 2026, and `@tiny-fish/sdk` 0.8.0 and `tinyfish` 0.9.0 on PyPI on 29 September (30). The TypeScript SDK had nine releases and the Python SDK seven since 22 July 2026 (20). The Pulse changelog stops at 22 June 2026. Support is by e-mail and Discord, and the public MCP proxy repository has commits to 8 October. We didn't read issue reply times (8 of 15 on the closed-service line). The hosted MCP server is in the official MCP registry as `ai.tinyfish.agent/web-agent`, version 1.0.1, published 17 February 2026, and both SDKs are current (15). The MCP proxy repository runs CI, Renovate, and OSV and secrets scanners, and publishes through npm trusted publishing. The SDK packages name no source repository (7).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: the trust centre at trust.tinyfish.ai is drawn by script (Vanta), so the ISO 27001 certificate, any SOC 2 report, DPA and subprocessor list were unread
- unchecked: the OpenAPI documents for Fetch, Search and Research. We read the main and Browser documents only, to keep requests to the API host few
- unchecked: whether the hosted MCP server's `tools/list` carries `readOnlyHint` or `destructiveHint`. Reading it needs a signed-in account
- unchecked: GitHub issue reply times, and whether the dashboard supports key expiry or per-key limits
- The Enterprise SLA of 99.99 per cent is a line on the pricing page. No SLA document was found
- The MCP docs page lists 17 tools while the MCP proxy repository's reference lists 28, with profile and monitor tools added
- The MCP `search` tool definition tells the model to prefer it over native web search tools and to use it when the user hasn't asked for a search. Recorded as a fact, with no deduction
- The privacy policy names TinyFish Inc. in Cupertino and the terms name Tiny Fish, Inc. in Los Altos. We used the name in the terms as the legal entity
- The status page reports downtime in whole hours per day, so short incidents may be rounded up
- The lead was right about the interfaces. It didn't mention the Monitor API, the CLI or MPP wallet top-ups, which the docs also describe

## Weaknesses

- The terms of 17 August 2026 allow Tiny Fish to train and improve its models on customer data. No opt-out was found in the reviewed pages
- The status page shows downtime of an hour or more on six days for the Browser API and five for the Search API between 11 July and 8 October 2026, with no incident write-ups
- No idempotency keys on run or session creation. The docs say retrying a failed request may create duplicate runs
- API keys have no scopes and no read-only mode was found. No prompt-injection guidance was found in the API docs
- No retention periods, public DPA or subprocessor list on the pages we could read. The trust centre is drawn by script and was unread
- The Pulse changelog's newest entry is 22 June 2026, and the Python and TypeScript SDKs are 0.x with no public source repository found

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Check `result` as well as `status`. A run can be `COMPLETED` while the goal failed, with `result.status` set to `failure`
- Don't retry `POST /v1/automation/run` blindly after a timeout. There are no idempotency keys, so list runs first to avoid duplicates
- Use `/v1/automation/run-async` or `/run-sse` when a run may need cancelling. Runs started with the synchronous endpoint can't be cancelled
- Allow 60 seconds for Browser session creation, connect Playwright to `cdp_url` (not `base_url`), and `DELETE` the session when done because billing is per minute
- Never put passwords in `goal`, which is logged. Pass `use_vault: true` with `credential_item_ids`, or `use_profile: true` for saved login state

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.