Paperclip

# Fix list: Paperclip

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/paperclip, the October 2026 research run, assessed 8 October 2026. Grade C, 59 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 Paperclip: 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, 66 out of 100, up to 6.8 more on the total

Why it scored 66: Read as local software. Official npm package `paperclipai` (Node.js 24.11 or newer) and a Docker image on GHCR (20). 2,429 test files and 20 workflows. The Actions page for master on 8 October 2026 showed recent runs passing, with two failed Runner Full-Stack E2E runs (681 and 682) before two that passed (22). 2,849 open issues, 125 labelled bug, 1,599 opened before 8 July, and 78 closed in the last 30 days against 686 opened (6). Calendar versions in semver syntax, with a Breaking Changes section and an upgrade guide in release notes, but breaking changes arrive in ordinary releases and 2026.916.0 carried five (10). No 1.0. The npm `latest` tag is called the stable channel and follows a beta soak of at least three days, while the roadmap says the product is still moving quickly, so half credit (8).

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, 64 out of 100, up to 6.3 more on the total

Why it scored 64: Harness reading of the framework checklist, as used for the other harnesses. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. Anonymous usage telemetry is on by default and goes to telemetry.paperclip.ing. The README lists what is excluded and four opt-outs, and Sentry and OpenTelemetry stay off until configured. We found no first-run notice in the CLI source (18). Board approvals for hires and strategy, execution policies and per-action connection permissions, but `claude_local` defaults `dangerouslySkipPermissions` to true, `codex_local` defaults to bypassing approvals and the sandbox, local agents run on the host, sandbox providers are optional plugins and the default `local_trusted` mode has no login (10). A documented `low_trust_review` preset contains agents that read untrusted pull requests or tickets. It is opt-in (10). An activity log of mutating actions, run logs, cost events and audited board API keys (14). SECURITY.md with GitHub private reporting and twelve advisories published, two with CVEs. No security.txt, bug bounty or SOC 2 report found (12).

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, 60 out of 100, up to 5 more on the total

Why it scored 60: No payment protocol (0). Read with the self-hosted rule. The software is free under MIT with nothing to buy (20). No card and no Paperclip account (20). One `npx` command installs and starts it with no signup (20). Paperclip Cloud is a waitlist with no published price, so there is no paid option to score.

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. Agent ergonomics, 70 out of 100, up to 4.9 more on the total

Why it scored 70: Harness reading of the framework line, scored on what an agent or pipeline driving it has to supply. The MCP server lists 44 tools with no toolsets or read-only subset found, offset by compact reads such as `paperclipInboxLite` and the heartbeat context (12). Budgets by company, agent and project with an automatic pause at 100 per cent, though the README says enforcement uses recorded spend and can lag work in flight (16). `onboard --yes`, CLI commands with `--json`, the REST API and routines started by cron, webhook or API (16). Atomic task checkout with a documented 409, `Idempotency-Key` on routine triggers, saved sessions and bounded recovery (17). One command starts a server with embedded PostgreSQL, but it needs Node.js 24.11 and each harness CLI installed and signed in on the host, and we found no client SDK beyond the TypeScript plugin SDK (9).

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.

## 5. Schema & documentation, 81 out of 100, up to 3.1 more on the total

Why it scored 81: The running server publishes OpenAPI at `/api/openapi.json` and validates requests with Zod, and the MCP server's tools take Zod schemas. No copy of the spec is published on the docs site, and paperclip.ing/openapi.json describes only the website (22). llms.txt at paperclip.ing, and docs pages answer as Markdown at `.md` (10). An API reference page per resource and task guides, but the MCP tool descriptions are one line each, such as List agents in a company (13). Typed inputs throughout, apart from the `paperclipApiRequest` tool, which takes any path under `/api` and a JSON body (12). cURL, JavaScript and Python examples and a status-code table, with one `{error}` shape and an occasional `details` field (12). A changelog page for every stable release and release notes in the repository, but the API path carries no version (12).

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, 65 out of 100, up to 3.1 more on the total

Made of editorial 73, provenance 56.

Why it scored 65: MIT (30). The privacy policy of 23 July 2026 separates self-hosted use from the hosted services and names Paperclip Labs, Inc., but keeps data as long as reasonably necessary, lists no subprocessors, and the 90-day telemetry retention appears only in source. The terms grant a broad licence over content sent to the services and allow model training on opt-in detailed telemetry (16). Release notes carry dated Breaking Changes sections and upgrade guides, with no written deprecation policy or notice period (10). The README discloses telemetry, what it excludes and four opt-outs, and the event contract is public in the repository, but it is on by default (17).

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: paperclip.ing, registered 2026-03-02 (under a year) (0 of 15)
- Terms of service: read, states 5 of the 7 things a reader expects (8.3 of 10)
- Privacy policy: read, states 7 of the 8 things a reader expects (9.3 of 10)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 78 out of 100, up to 1.9 more on the total

Why it scored 78: Version 2026.1005.0, dated 5 October 2026 in its notes and published on GitHub and npm on 6 October (30). Eleven stable releases since 20 July 2026 (20). 650 pull requests merged in the last 30 days, but 2,849 issues and 3,338 pull requests are open, 1,396 open issues have no comment, and 309 of the 472 still-open issues from the last two weeks have none (8). `paperclipai` and `@paperclipai/mcp-server` are both at 2026.1005.0 on npm. We did not check the MCP registry (12). CI on master, a lockfile refresh workflow and signed image attestations (8).

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.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 2026-04-10 to 2026-04-16. Ten advisories published in a week, four of them critical, among them GHSA-68qg-g8mg-6pr7 (CVE-2026-41679, 10.0, unauthenticated remote code execution through an import authorisation bypass) and two cross-tenant agent key flaws rated 9.9. All ten list 2026.416.0 as the fix. Fixed and published, a little under six months old, -5. https://github.com/paperclipai/paperclip/security/advisories
- 2026-07-22. GHSA-x8hx-rhr2-9rf7 (9.6), drive-by remote code execution against the default `local_trusted` mode through DNS rebinding. The advisory names no patched version. The current source applies the hostname guard to `local_trusted` private deployments, so we read it as fixed without a documented version, -3. https://github.com/paperclipai/paperclip/security/advisories/GHSA-x8hx-rhr2-9rf7
- 2026-04-16. GHSA-gqqj-85qm-8qhf (8.7), a `codex_local` agent inherited a Gmail connector from the owner's ChatGPT account and sent real email. The advisory names no patched version, and `codex_local` still defaults to bypassing approvals and the sandbox, -2. https://github.com/paperclipai/paperclip/security/advisories/GHSA-gqqj-85qm-8qhf

## 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 GitHub API rate limit stopped us reading ten of the twelve advisories in full, so their severity and fixed versions come from the advisory list
- unchecked: whether Paperclip is in the official MCP registry
- unchecked: reply times on issues. The counts come from the GitHub search API
- GHSA-x8hx-rhr2-9rf7 and GHSA-gqqj-85qm-8qhf name no patched version. We read the first as fixed from the current source and found no change of default for the second
- We found no first-run telemetry notice in the CLI source, and did not install or run the software
- No price, date or terms for Paperclip Cloud beyond the waitlist page
- The licence file in the repository names Paperclip AI and the terms name Paperclip Labs, Inc.
- The scout gave 6 October 2026 for the latest release. The release notes and the site say 5 October, and GitHub and npm published it on 6 October
- Judgement call. The listing is in agent harnesses, but Paperclip drives other harnesses and does not run the model loop for most adapters, so `agent.harness` is left out of its capabilities

## Weaknesses

- `claude_local` defaults `dangerouslySkipPermissions` to true and `codex_local` defaults to bypassing approvals and the sandbox, with agents running on the host
- Twelve published security advisories since April 2026, five critical, including CVE-2026-41679 (unauthenticated remote code execution, fixed in 2026.416.0)
- Anonymous usage telemetry is on by default and goes to telemetry.paperclip.ing until an opt-out is set
- 2,849 open issues on 8 October 2026, 1,396 of them with no comment, and 78 closed in 30 days against 686 opened
- Calendar versions with no 1.0, and release 2026.916.0 carried five breaking changes

## 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.

- Set `PAPERCLIP_TELEMETRY_DISABLED=1` or `DO_NOT_TRACK=1` before the first start. Telemetry is on by default
- Set `dangerouslySkipPermissions` and `dangerouslyBypassApprovalsAndSandbox` to false on agents that read untrusted input, or run them in a sandbox provider
- Install with Node.js 24.11 or newer, and install and sign in to each harness CLI on the host first. Paperclip assumes they are there
- Use `--bind lan` or `--bind tailnet` at onboarding for anything beyond one machine. The default `local_trusted` mode treats every request as the board admin
- Treat 409 on task checkout as owned by another agent and pick different work. The API docs say not to retry

## 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.