# tex-mcp-web (pypi · tex-mcp-web)

Comment on a LaTeX paper's PDF and have Claude Code or Codex edit the source, with Overleaf sync.

- Trust score: 64/100 (medium)
- Change this week: −11
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-11-21

## Components

- pypi · `tex-mcp-web`: 64/100 (this document), [markdown](https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web.md), [page](https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web)

## Channel facts

- Registry: `pypi`
- Package: `tex-mcp-web`
- Version: `0.12.0`
- Transport: `stdio`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically from public evidence about the published package, including repeated runs of it in an isolated sandbox, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-11-21.

- **Supply Chain Security**: 50/100
  - Malware scan not yet available for this package.
  - No known CVEs affecting this package version or its production dependencies.
  - Runs hatchling.build at install time, a recognised build step with no custom scripting around it.
  - 3 of 49 dependencies flagged as unhealthy.
- **Provenance & Transparency**: 97/100
  - Source repository is publicly reachable at the declared URL.
  - Cryptographically verified build provenance (signed, bound to MiiKiyoshi/tex-mcp-web).
  - Clear OSI-approved license (MIT).
  - Actively maintained (last published 0 days ago).
  - Disclosure check failed: no security disclosure policy was found in the source repository.
- **Schema Quality & AI Usability**: 52/100
  - AI-judged instruction clarity (good).
  - Context-footprint check failed: tool/resource definitions use about 2726 tokens (~340/item across 8 items; 8 tools + 0 resources), over budget; trim descriptions and params.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 27/100
  - Stability observed for 8 of 30 days with no destabilising changes; credit accrues until the full window elapses.
- **Tool Coverage**: 77/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 32% of tool parameters carry a description.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - We read all 8 captured tool definition(s), and no name or description among them implies an irreversible operation.
  - An AI judge read all 9 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 100/100
  - Implements a supported MCP spec version (2025-11-25); the latest is 2026-07-28.

## Install

### How do I install the tex-mcp-web server?

tex-mcp-web runs locally as a PyPI package, launched with uvx tex-mcp-web mcp. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add miikiyoshi-tex-mcp-web -- uvx tex-mcp-web mcp
```

### Cursor

```json
{
  "mcpServers": {
    "miikiyoshi-tex-mcp-web": {
      "command": "uvx",
      "args": [
        "tex-mcp-web",
        "mcp"
      ]
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "miikiyoshi-tex-mcp-web": {
      "command": "uvx",
      "args": [
        "tex-mcp-web",
        "mcp"
      ]
    }
  }
}
```

### Codex

```bash
codex mcp add miikiyoshi-tex-mcp-web -- uvx tex-mcp-web mcp
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "miikiyoshi-tex-mcp-web": {
      "type": "local",
      "command": [
        "uvx",
        "tex-mcp-web",
        "mcp"
      ],
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add miikiyoshi-tex-mcp-web --command uvx --arg tex-mcp-web --arg mcp
```

### Hermes

```yaml
mcp_servers:
  miikiyoshi-tex-mcp-web:
    command: "uvx"
    args: ["tex-mcp-web", "mcp"]
```

### Netclaw

```json
{
  "McpServers": {
    "miikiyoshi-tex-mcp-web": {
      "Transport": "stdio",
      "Command": "uvx",
      "Arguments": [
        "tex-mcp-web",
        "mcp"
      ]
    }
  }
}
```

### Vellum

```bash
assistant mcp add miikiyoshi-tex-mcp-web -t stdio -c uvx -a tex-mcp-web mcp
```

### Other

```json
{
  "mcpServers": {
    "miikiyoshi-tex-mcp-web": {
      "command": "uvx",
      "args": [
        "tex-mcp-web",
        "mcp"
      ]
    }
  }
}
```

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-11-21 (score 64, +1)

- [functional] Package version: 0.11.4 → 0.12.0

### 2026-11-20 (score 63, 0)

- [security improvement] Malware scan: unverified → pass
- [functional] Package version: 0.11.3 → 0.11.4

### 2026-11-19 (score 63, +1)

- [security improvement] Malware scan: unverified → pass
- [functional] Package version: 0.11.1 → 0.11.3
- [functional] Package version: 0.11.1 → 0.11.2

### 2026-11-18 (score 62, −3)

- [security regression] Malware scan: pass → unverified
- [security improvement] Known CVEs: unverified → pass
- [security improvement] Known CVEs: unverified → partial
- [functional improvement] Dependency health: unverified → 0.98
- [functional improvement] Dependency health: unverified → partial
- [functional] Package version: 0.10.1 → 0.11.1
- [functional] Package version: 0.10.1 → 0.11.0

### 2026-11-17 (score 65, +4)

- [security regression] Known CVEs: pass → unverified
- [security improvement] Malware scan: unverified → pass
- [functional regression] Dependency health: 0.98 → unverified
- [functional] Package version: 0.10.0 → 0.10.1

### 2026-11-16 (score 61, −15)

- [security regression] Malware scan: pass → unverified
- [functional regression] Schema quality: 280 → 317
- [functional regression] Schema quality: 280 → 316
- [functional regression] Schema quality: 280 → 308
- [functional] Package version: 0.9.1 → 0.10.0
- [functional] Package version: 0.9.1 → 0.9.7
- [functional] Package version: 0.9.1 → 0.9.6

### 2026-11-15 (score 76, +1)

- [security regression] Malware scan: pass → unverified
- [functional regression] Schema quality: 241 → 280
- [functional] Package version: 0.8.5 → 0.9.1
- [functional] Package version: 0.8.5 → 0.9.0

### 2026-11-14 (score 75, +15)

- [security improvement] Malware scan: unverified → pass
- [functional improvement] Stability: unverified → 0.03
- [functional] Package version: 0.8.4 → 0.8.5

## MCP tools (8)

### `guide` (~51 tokens)

Paths of the files to read in full before editing the paper: the paper's guideline and the guidelines it extends, base first, when guideline is set, and the terms its reference papers use, when references is set.

### `read_comments` (~999 tokens)

Read review threads: what is new, selected threads in full, or a listing. ``new`` answers a Call agent press in one call. For each open thread it returns what the reviewer has written that no earlier call with ``new`` returned and no answer has covered yet, with the thread's rev, which a write passes back. A thread's last answer covers everything before it; an answer is your reply, your Overleaf draft, or a public reply the reviewer sent to Overleaf. A suggestion the reviewer applied is returned once, as an entry marked applied with the lines it changed. Your wording is then in the file, and the entry needs no reply, except on a thread from Overleaf with no draft, where the change calls for a reply. A suggestion or an Overleaf draft the reviewer rejected is also returned once, as an entry marked rejected with the rejected text, and it stays in the thread for a later read by id. It needs no reply. Do not propose the same thing again unless the reviewer asks; if they give a reason, it is in one of their replies. The answer gives the moment the read started from (``from``) and the one it moved to (``cursor``), so a turn that went wrong can be repeated with since=<from>; to see again what was already answered, read the thread by id. A thread from Overleaf is marked on_overleaf, and one the reviewer started for Overleaf but has not sent there yet is marked new_thread; a thread with a draft for Overleaf comes with that draft. An entry is public, on Overleaf for everyone in the project, when it has ``name`` (written there) or ``overleaf`` (sent there from the page), and so is ``comment``, the first message, of a thread marked on_overleaf. Every other entry is private, for the reviewer and you, and so is everything in a thread that is not on Overleaf. ``ids`` returns those threads in full, for when you no longer remember a conversation; with ``since`` it returns only the entries after that time. With neither, it returns a listing: the newest ``limit`` threads, each request…

Input parameters:

- `file`: A source file's name or path, like 'intro.tex' or 'tex/intro.tex', to keep only the threads on that file. For a listing or a search.
- `find`: Words to search for, ignoring case, in a thread's entries and its draft for Overleaf.
- `ids`
- `limit` (integer)
- `new` (boolean): Only what the reviewer has written since the last call with new; cannot be combined with ids.
- `on_overleaf`: true keeps only threads from Overleaf or started for it, false only the page's own threads, which are entirely private, and leaving it out keeps both. For a listing or a search.
- `save` (boolean)
- `since`: A time this server printed, like '09-19 00:52'; only what came after it.
- `status` (string)
- `unanswered` (boolean): Only threads where the reviewer wrote to you after the thread's last answer: your reply, your Overleaf draft, or a reply sent to Overleaf.
- `until`: A time this server printed; a search covers only what came up to the end of that minute.

### `overleaf` (~179 tokens)

Compare this folder with its Overleaf project. state reads Overleaf again and reports only what differs: get (Overleaf's edits not here), send (edits here not on Overleaf), choose (files waiting for the reviewer to choose), and, in a git clone, the commits behind and ahead. suggested lists the edits made on Overleaf with track changes on, which wait for the reviewer to accept or reject them on the page; Overleaf's text contains the words of a suggested insertion but not those of a suggested deletion. get brings Overleaf's edits into this folder (git pull in a git clone, Get otherwise); it changes only files here, and a conflict changes nothing. push sends a git clone's commits (git push). Without git, only the reviewer sends edits, from the page.

Input parameters:

- `action` (string)

### `compile` (~219 tokens)

Compile the paper. Returns whether it succeeded, which pages changed, where the log is, and, when it failed, where each error is. Call once after a batch of source edits. ``pages_changed`` compares extracted PDF text and excludes visual-only changes. Read the log yourself for warnings and for the context around an error. A successful build also reports how the paper fits a page limit, as the page's Problems tab does. ``body_end`` is the last line before the bibliography, with its page, its column and the number of columns, the space below it to the end of the text block in points and in lines (once the paper has full columns to measure by), and its source. ``short_lines`` lists the paragraphs whose last line is narrower than a quarter of its column, including those in a list item or in centered text such as a table's caption, each with its page, column, words, width and source; shortening such a paragraph removes that line. A source is the line that contains a paragraph's last words.

### `write_comments` (~671 tokens)

Write to a comment thread. What you write is private, on the review page for the reviewer and you. overleaf drafts what becomes public, on Overleaf for everyone in the project, once the reviewer sends it. add: text and anchor, plus changes, quoted as for suggest, when the reviewer has asked for proposals left as comments; the new thread then carries that proposal. reply: id and text, or a saved draft on its own. suggest: id, text, changes, rev. Each change's old is quoted from the file the comment is on and must occur there exactly once, as an editing tool expects; never give a line or a column. It replaces the thread's single current proposal and adds text as a reply, so the conversation stays complete. It is refused on a thread the reviewer has not written in: propose on their thread instead of opening your own beside it. withdraw: id, text, rev. overleaf: id, text, rev, and edits naming the source ranges the reply describes, by their current lines; later edits move these ranges as they move a comment, so the draft does not need rewriting when lines shift. On a thread that came from Overleaf (on_overleaf), it drafts the public reply that the reviewer may send there. A thread has one draft, which this replaces, and only the reviewer sends it. A first message the reviewer wrote themselves (overleaf_draft_by: reviewer) is not replaced: your text is added to the thread as your private reply suggesting the wording. The reviewer is asked first when a file that contains those ranges differs between here and Overleaf. Since the draft is public once sent, word it for the people on Overleaf, and answer what the reviewer asks you in the thread with reply, which is private. With anchor (a source_range) instead of id and rev, overleaf drafts a new public thread for Overleaf on those lines, with text as its first message. The result is a thread that has the draft and is not on Overleaf. Call overleaf with its id and rev to redraft it. Only the reviewer sends it, and only once…

Input parameters:

- `action` (string, required)
- `anchor`
- `changes`
- `draft`
- `edits`
- `entry`
- `id`
- `rev`
- `text`

### `image` (~129 tokens)

Render one PDF target as PNG. Choose exactly one of page (with optional bbox), source="file.tex:lstart-lend", or comment. Margin expands an exact bbox only while rendering. A multi-page source range renders the page with the most SyncTeX matches.

Input parameters:

- `bbox`
- `comment`
- `dpi` (integer): 96 is enough to read text. 150 or more shows fine detail and costs more tokens.
- `grayscale` (boolean): Set false when colour itself is being checked.
- `margin` (number)
- `page`
- `source`

### `listen` (~51 tokens)

Return a command and client-specific instructions for listening for Call agent presses. Run the returned command as its how field describes; how is written for the connected client. Stop it when the review is finished or the user asks.

### `setup_info` (~62 tokens)

Report setup paths, the existing binding with the paper's folder it serves, the guidelines to choose from, the path of the tex-mcp-web command, whether translation is set up, and whether the reviewer's Overleaf cookie is saved, without connecting or changing files.

## Diagnostics

Captured diagnostic sections: Provenance, Install scripts, Dependencies. The full working is on the page: https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web#diagnostics

## Score history

- 2026-11-21: 64
- 2026-11-20: 63
- 2026-11-19: 63
- 2026-11-18: 62
- 2026-11-17: 65
- 2026-11-16: 61
- 2026-11-15: 76
- 2026-11-14: 75
- 2026-11-13: 60

## Common questions

### What is the tex-mcp-web server?

tex-mcp-web is listed in the public MCP registry as io.github.MiiKiyoshi/tex-mcp-web. Comment on a LaTeX paper's PDF and have Claude Code or Codex edit the source, with Overleaf sync. This page covers its PyPI package (tex-mcp-web).

### Is the tex-mcp-web server safe to use?

tex-mcp-web scores 64 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 21 November 2026. Its build provenance is signed and verified. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the tex-mcp-web server expose?

tex-mcp-web exposes 8 tools: guide, read_comments, overleaf, compile, write_comments, and 3 more. Their descriptions and schemas cost roughly 2,361 tokens of context every time the server is loaded.

### Is the tex-mcp-web server still maintained?

tex-mcp-web is still listed as active in the MCP registry. We last reached this channel on 21 November 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

### What licence is the tex-mcp-web server under?

tex-mcp-web declares the MIT licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

### What did VerifyMCP test on the tex-mcp-web server?

VerifyMCP ran 15 checks on tex-mcp-web, covering Known CVEs, Install scripts, Dependency health, Source repository and 11 more. One item could not be completed, a gap on our side rather than a finding about tex-mcp-web: Malware scan. Endpoint Security and Transport & Reachability do not apply to a PyPI package, so they were not tested. The score does not measure everything, and the scoring documentation lists what it leaves out.

## Links

- PyPI project: https://pypi.org/project/tex-mcp-web/
- Socket report: https://socket.dev/pypi/package/tex-mcp-web
- Repository: https://github.com/MiiKiyoshi/tex-mcp-web
- Changelog RSS feed: https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web.xml
- Changelog JSON feed: https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web/changelog.json
- Trust report (JSON): https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web.json
- HTML version of this page: https://verifymcp.io/servers/miikiyoshi-tex-mcp-web/tex-mcp-web
