Thunder Compute

# Fix list: Thunder Compute

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/thunder-compute, the October 2026 research run, assessed 9 October 2026. Grade C, 56.1 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 Thunder Compute: 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. Payments & pricing, 20 out of 100, up to 10 more on the total

Why it scored 20: No x402, MPP or L402 in the docs, the OpenAPI document or the pricing page (0). Per-GPU-hour, per-vCPU-hour and storage prices are public on the pricing page and from `GET /v2/pricing` without a token (20). No general free tier or trial. A $20 credit is limited to students with a US university email, and the MCP quickstart asks for a payment method before connecting (0). Signup and the first OAuth consent happen in a browser. `create_token` issues API tokens only after that sign-in (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.

## 2. Security & auth, 51 out of 100, up to 8.6 more on the total

Why it scored 51: The MCP server uses OAuth 2.0 with discovery, refresh tokens and ten scopes that separate read from write for instances, snapshots and sandboxes. The REST API uses named API tokens that are created and revoked one at a time, with no scopes found. No secret travels in a URL (25 of 30). Read scopes exist, and the CLI requires `--yes` to delete without a prompt. No confirmation step for `delete_instance` or `delete_snapshot` was found in the MCP docs, and both are described as irreversible (8 of 20). Responses are infrastructure metadata, except `run_command` output and sandbox logs, with no guidance found on untrusted output (8 of 15). `GET /tokens` returns creation and last-used times and sandbox logs are listed by API. No audit log was found (4 of 15). No security.txt, disclosure policy or bug bounty was found. `llms.txt` states SOC 2 Type II and GDPR compliance with no report or trust page linked, and the DPA promises an audit summary on written request and incident notice within 72 hours (6 of 20).

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. Agent ergonomics, 48 out of 100, up to 8.5 more on the total

Why it scored 48: No field selection, and `GET /instances/list` returns every instance in one map. The MCP server has 28 tools in one set, with no toolsets found (12 of 25). `limit` up to 100, `page_token` and a status filter on sandboxes and sandbox logs. No pagination on instances, snapshots or keys (8 of 20). Errors carry a numeric `code`, a string `error` such as `invalid_request` and a `message`, documented per operation (14 of 20). Duplicate SSH keys are a no-op and port changes are deduplicated in one transaction, but create calls have no idempotency key. MCP tool annotations were not read (7 of 20). A create call needs five fields. The official Python SDK covers sandboxes only, and the Go CLI covers instances (7 of 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.

## 4. Reliability, 65 out of 100, up to 7 more on the total

Why it scored 65: Graded on the hosted lines for the REST API and the hosted MCP server. Status page on incident.io at status.thundercompute.com with Website, API and Instances components and monthly history (20). No incidents listed for July to October 2026, and each component shows 100 per cent (30). No rate limits with numbers were found in the docs or the OpenAPI document (0). No 429 or retry guidance in the docs. The CLI source reads a `Retry-After` header on errors, adding a duplicate SSH key is a no-op and port updates are atomic, but instance creation has no idempotency key (5 of 15). No SLA found, and the terms say availability is not guaranteed (0). The API and MCP server carry no beta label. Sandboxes need access enabled for each organisation (10).

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.

## 5. Schema & documentation, 70 out of 100, up to 4.9 more on the total

Why it scored 70: Public OpenAPI 3.1 document at api.thundercompute.com:8443/openapi.json with 41 operations and 72 schemas (25). `llms.txt`, `llms-full.txt` and a Markdown copy of every docs page (10). 28 of 41 operations have a description. Instance operations state side effects such as the private key being returned once, while most sandbox operations have a summary only (12 of 20). Required fields are marked, but the document has six enums, `gpu_type` and `template` are free strings and the instance list is a map keyed by ID (8 of 15). 12 examples and one error schema with `code`, `error` and `message`. Most operations document 400, 401 and 500, and none documents 429 (9 of 15). `/v1` and `/v2` paths, with legacy endpoints and the `mode` field marked as retained or deprecated. No API changelog was found. The CLI has a dated one (6 of 15). The MCP tool definitions need a signed-in session and were not read.

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.

## 6. Transparency & trust, 64 out of 100, up to 3.2 more on the total

Made of editorial 51, provenance 77.

Why it scored 64: Closed service under Terms and Conditions dated 28 September 2026 and Delaware law. The CLI is MIT. The OpenAPI document names Apache 2.0 and a placeholder terms address (15 of 30). The privacy policy of 28 September 2026 says instance data is not used to train AI models, the terms say instance data may be deleted 60 days after the account was last active, and a DPA is published. The sub-processor list names Firebase for sign-in while the MCP docs name Stytch, and the FAQ says instances are EC2 while the list names Azure (18 of 30). No deprecation policy. The terms allow changes without notice, and the billing docs say rates can change without notice. Legacy endpoints are marked as retained for compatibility (6 of 20). Five sub-processors with locations are listed and the DPA sets 10 business days' notice of changes. The list is dated 27 August 2025 and omits Stytch and Sentry, which the MCP docs and the CLI source name (12 of 20).

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: thundercompute.com, registered 2024-02-29 (2 years) (7 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 2 clauses that cost points (5.1 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 79 out of 100, up to 1.8 more on the total

Why it scored 79: The newest CLI release is v2.1.0 of 16 September 2026, and the terms and privacy policy were updated on 28 September 2026 (30). Four CLI releases in the last 90 days, on 15 July, 30 July, 15 September and 16 September 2026 (20). A dated CLI changelog, a support address and a Discord server the docs say the founders answer. No API changelog, and the CLI repository shows no open issues (10 of 15). Listed in the official MCP registry as `io.github.Thunder-Compute/thunder-compute`, version 1.0.0 from 9 April 2026, with a remote address that differs from the one in the docs. The CLI and the `thunder-sandbox` SDK are current (13 of 15). The CLI repository has tests, a lint configuration and release checksum verification. No public CI workflow was found in it (6 of 10).

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.

- The lead was right about the interfaces. It did not mention sandboxes, which are a second product on the same API and need access enabled for each organisation.
- unchecked: the MCP tool definitions, input schemas and annotations. `tools/list` answers 401 without a signed-in session, so the 28 tools are counted from the docs table.
- unchecked: the PyPI pages for `thunder-sandbox` and `tnr`. The project page needs scripts and PyPI's robots.txt disallows the JSON API, so versions, dates and downloads were not read.
- unchecked: the SOC 2 Type II and GDPR claims in `llms.txt`. No report, auditor or trust page was found on the site.
- The docs give the MCP address as `https://www.thundercompute.com/mcp`. The official registry entry of 9 April 2026 gives `https://api.thundercompute.com:8443/mcp`. We did not test the second.
- `llms.txt` has sections titled Preferred Summary Style for AI and Additional Guidance for AI that tell models how to describe the company. Recorded as a fact. We did not act on them and took no deduction.
- The CLI source initialises Sentry error reporting when a DSN is built in. No opt-out or mention was found in the docs, and Sentry is not on the sub-processor list. The CLI is not the surface graded, so no deduction was taken.
- The FAQ says instances are EC2 instances, the sub-processor list names Azure, and the technical specifications say only North America. The hosting provider and regions are not established.
- `llms.txt` describes prototyping and production modes. The OpenAPI document marks `mode` as deprecated and the pricing page no longer shows modes.
- Whether a user can grant only read scopes at the OAuth consent screen is unknown.

## Weaknesses

- No rate limits, 429 guidance or SLA were found in the reviewed documentation, and the terms disclaim availability
- `POST /instances/create` has no idempotency key, so a retried launch can start a second billed instance
- Instances cannot be stopped. Pausing means taking a snapshot, deleting the instance and restoring later
- No general free tier. The $20 credit is limited to students with a US university email
- No security.txt or disclosure policy was found, and the SOC 2 Type II claim in `llms.txt` has no report or trust page linked
- The sub-processor list, last updated 27 August 2025, names Firebase for sign-in while the MCP docs name Stytch

## 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.

- Call `GET /v2/status` or the `get_availability` tool before creating an instance. Availability can change before launch, and creation fails when a type is sold out.
- List instances before retrying a failed create, because the call has no idempotency key.
- Pass `public_key` on create. If omitted, the response carries a generated private key that is returned once.
- To pause work, create a snapshot, delete the instance and later create a new instance from the snapshot. Snapshot storage keeps billing until deleted.
- For headless use set `TNR_API_TOKEN` to a token from the console. The MCP server needs a browser sign-in on first connection.

## 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.