Graphify

# Fix list: Graphify

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/graphify, the October 2026 research run, assessed 26 October 2026. Grade B, 69 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 Graphify: 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, 58 out of 100, up to 8.4 more on the total

Why it scored 58: Read with the local-software lines, since the owner runs it. `graphifyy` installs from PyPI with uv or pipx, with `requires-python >=3.10` stated (20). The CI workflow runs pytest on Python 3.10, 3.12, 3.13 and 3.14, an end-to-end `graphify install`, the Pi guard tests and generated-skill checks on every push and pull request to v8, over 458 test files. Whether the latest runs pass couldn't be read, because GitHub's web pages weren't reachable (15 of 25). Open issues weren't read. The changelog fixes many extraction regressions each release, each credited to a reporter and a pull request, which reads as handled rather than left open (15 of 25). CHANGELOG.md marks breaking changes, such as full-path node IDs in 0.9.0 and Java `inherits` edges in 0.8.18, and announces AST cache bumps, but on 0.x they ship in ordinary releases with no Breaking section (8 of 15). Version 0.9.95 (0).

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, 66 out of 100, up to 6 more on the total

Why it scored 66: Local-software reading, as for vibe-trading. 30 for what leaves the machine by default, 20 for approvals and least privilege, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. Code parsing is local and no telemetry code was found. The HTTP server binds 127.0.0.1 with a Host allow-list and a constant-time key check, but only warns when bound to 0.0.0.0 without a key, and turns DNS-rebinding protection off for a wildcard bind. Headless `extract` sends docs to whichever provider's key is in the environment, with a printed notice, and community naming falls back to `claude -p` when no key is set (21). No MCP tool writes, and `uninstall --purge` refuses to delete the project or home folder. The Claude hooks inject a MANDATORY instruction before every Bash, Grep, Read and Glob call once a graph exists, the default install edits the user-wide ~/.claude/CLAUDE.md, and git hooks rebuild on every commit and checkout. Settings files get a `.graphify-bak` copy (14). The semantic pass wraps each file in a hash-stamped `untrusted_source` block, defangs chat-template tokens, sanitises labels in MCP output and redacts credential-shaped text, and SECURITY.md says this lowers injection rather than stopping it (11). An opt-in query log, a stderr line naming the backend and the variable that chose it, and a local cost.json (8). SECURITY.md with private reporting and a threat table, a valid security.txt, one published advisory with a fix, PyPI trusted publishing, and bandit and pip-audit in CI set not to fail the build. The SOC 2 engagement covers the hosted product only and has no report yet. No bug bounty (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: Read with the self-hosted rule. No payment protocol (0). The package is free under Apache 2.0, needs no account or card, and builds a code graph with no key, so 20, 20 and 20 on the last three lines. Graphify Cloud, a separate hosted product with public plan prices, is not scored here.

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. Schema & documentation, 77 out of 100, up to 3.7 more on the total

Why it scored 77: MCP reading. All ten tools carry typed JSON Schema inputs in serve.py, though `get_node` and `get_neighbors` mark no field required and no tool has an output schema (20 of 25). llms.txt at docs.graphify.com, a Markdown copy of every page, llms-full.txt, and an MCP tool reference generated from serve.py (10). Descriptions say what each tool does in one sentence; the three PR tools also say when to call them, the graph tools don't say when not to (10 of 20). Types, defaults and a `bfs`/`dfs` enum, but `depth` is limited to 1 to 6 in text and code rather than in the schema, and `relation_filter` is a free string (10 of 15). Worked examples in the README and docs, documented exit codes 0, 3 and 4, refusals that list candidate nodes, and a troubleshooting page (12 of 15). A dated CHANGELOG.md with 238 entries, a docs changelog, and `schema_version` in graph.json (15).

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.

## 5. Agent ergonomics, 77 out of 100, up to 3.7 more on the total

Why it scored 77: MCP reading. Ten compact tools, about 7,900 characters of tool-definition source in serve.py (25). `token_budget` on `query_graph`, `get_neighbors` and `get_community` (default 2,000), `depth` capped at 6, `max_hops`, `top_n`, `relation_filter` and `context_filter`, and `--budget` on the CLI. No cursor paging (17 of 20). Exit codes for answer, ambiguous name and no match, ambiguous names refused with candidates instead of guessed, and a clear message when graph.json is missing or corrupt. MCP errors come back as text (15 of 20). Every tool reads, so calls are safe to repeat, but none carries readOnlyHint or destructiveHint (10 of 20). `graphify install` and `/graphify .` build a first graph, `graph_stats` is the documented first call, and code needs no key. Python only, though the CLI works from any language (10 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.

## 6. Maintenance & community, 81 out of 100, up to 1.7 more on the total

Why it scored 81: 0.9.95 was published on 23 October 2026, three days before this check (30). 67 PyPI versions in the 90 days to 26 October, 0.9.29 to 0.9.95 (20). The 0.9.95 entry credits 21 community pull requests and the changelog names reporters on most fixes, with issue numbers past #4700. Issue threads and reply times weren't read, a judgement call (18 of 25). The official MCP registry holds `io.github.Graphify-Labs/graphify`, but that entry is Graphify Cloud's hosted server; the local `graphify-mcp` isn't listed. Published on PyPI (5 of 15). A frozen uv.lock in CI on four Python versions, dependency floors raised for named CVEs (starlette 1.3.1), and trusted publishing. No Dependabot configuration, and the audit jobs don't block (7 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.

## 7. Transparency & trust, 83 out of 100, up to 1.5 more on the total

Made of editorial 78, provenance 88.

Why it scored 83: The editorial half. Apache 2.0, with the MIT portion before 0.9.25 kept in LICENSE-MIT (30). The README's privacy section says what stays local, which env vars send docs to which provider, that Kimi routes to China and how to turn backends off, and the docs agree. The README still says every query is logged by default while the code and configuration docs say logging is off by default, and graphify.com's security page says the core makes no network calls of its own, which the `claude` CLI fallback for community names stretches (20 of 30). The changelog dates breaking changes and removals, SECURITY.md supports the latest 0.9.x only, and graph.json carries a schema version. No deprecation policy or notice period (8 of 20). No telemetry, stated in the README and on the security page, and none found in the source (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):

- Status page: not found (0 of 10)

## Deductions

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

- 2026-10-19. GHSA-pcc4-rvhr-2pr8 (moderate). In 0.7.2 to 0.9.69 the Fortran extractor's `cpp` step followed absolute and traversing `#include` lines in an untrusted `.F` file, reading host files into graph.json, the report and the model context on the default offline path. Fixed in 0.9.70 of 27 September 2026, with a Security line in the changelog the same day and the advisory published three weeks later. Fixed and documented, so -1. https://api.osv.dev/v1/vulns/GHSA-pcc4-rvhr-2pr8

## 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.

- Category. code, because the main job is reading a repository's structure for a coding agent, beside sourcegraph-mcp and rea. agent-memory was the alternative, since the graph persists between sessions and Graphicle sits there, but Graphify keeps no record of decisions or past sessions beyond an opt-in Q&A memory folder. Judgement call.
- Kind is `mcp`, as for rea and Graphicle, though the installed skill and CLI are the surfaces most assistants use. Graded with the local-software Reliability lines, the MCP lines for Schema and Ergonomics, the local-software reading of Security and the self-hosted Payments rule.
- firstReleased is 4 April 2026, the date of 0.1.1, the earliest version on PyPI. CHANGELOG.md has a 0.1.0 entry dated 3 April 2026 that PyPI doesn't list.
- unchecked: github.com/robots.txt. The shell's proxy answered the request with an API error and WebFetch's permission request timed out, so no GitHub web page was fetched and the repository was read from a `git clone --depth 1` instead.
- unchecked: the GitHub star count, CI run results, issue and pull request threads and the GitHub release pages. The lead's 135K matches the 135.4k in graphify.com's header on 26 October 2026 (135.3k on the terms page minutes later). YC's profile says 116K+ and its launch post 108K+. All three are the vendor's figures and none is used as popularity.
- unchecked: PyPI download counts. PyPI's robots.txt disallows `/pypi/*/json`. The YC profile's 6.5M+ downloads is the vendor's claim.
- The lead's about 40 languages matches the README's own count and the extension table in detect.py. graphify.com's security page says 36 bundled grammars. The lead's 20 and more assistants matches the README's table of 24 install targets.
- Benchmark figures in the README and BENCHMARKS.md (LOCOMO recall@10 of 0.497 against mem0 and supermemory, 82 per cent key-fact coverage), Graphify Cloud's 80 per cent token saving and the YC post's first to cross 100K stars mid-batch are Graphify Labs' claims, not checked and not deducted.
- The privacy policy says Graphify Labs maintains a SOC 2 Type II programme, while the security page says the engagement has started with no report. Both concern the hosted product. Recorded, not deducted.
- Deduction of 1 for GHSA-pcc4-rvhr-2pr8, fixed in 0.9.70 and disclosed in the changelog the day of the fix, weighed below vibe-trading's critical advisories. Judgement call.
- The claude CLI fallback for community names and the auto-detection of a backend from ambient keys are recorded as weaknesses, not deductions, because both print a notice and are documented in the code and the README's privacy section. Judgement call.
- Graphify Cloud (app.graphify.com and the OAuth MCP server at api.graphify.com/mcp) is a separate hosted product and wasn't graded. It could take its own listing.
- Graphicle. Its listing says its code-graph engine is a Rust port of Graphify. A grep of data/tools/graphicle.json for that line also printed part of its transparency note, which wasn't used.
- Our own products. Graphify doesn't do LocalGhost's job (a personal AI server over the owner's own data) or letme's (choosing tools from our grades). Graded the normal way.
- Nothing was installed or run. Tool counts, defaults and behaviour come from reading the source at 0.9.95.

## Weaknesses

- Headless extraction sends docs to whichever cloud provider's key is set, Gemini first and Kimi in China second, unless `GRAPHIFY_BACKEND` is set
- Community naming falls back to the `claude` CLI when no key is set, which sends node names to Anthropic through the owner's plan
- Version 0.9.95, with breaking changes such as node IDs in 0.9.0 shipped inside 0.x releases
- One moderate advisory, a host-file read through the Fortran extractor in 0.7.2 to 0.9.69, fixed in 0.9.70
- No readOnlyHint or destructiveHint annotations, and the local server isn't in the official MCP registry

## 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.

- Install `graphifyy` with two y's; other `graphify*` packages on PyPI are not the project's
- Set `GRAPHIFY_BACKEND=none` or use `--code-only` when no source or doc text may leave the machine
- Pass an absolute path to graph.json in the MCP config, and run `graph_stats` first to check the graph loaded
- Set `GRAPHIFY_API_KEY` before binding the HTTP server to 0.0.0.0; it only warns when no key is set
- Run 0.9.70 or later before indexing an untrusted repository, because earlier versions carry the Fortran file-read advisory

## 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.