Keystone is an MCP server published by io.github.Srikarmk, hosted at keystone-research.vercel.app. When last checked on 2026-11-22 it was working: it answers and lists its tools. It declares 3 tools, including search_library, read_paper, list_disagreements.
What an arXiv paper builds on, disputes and assumes, read verbatim from its LaTeX. No LLM.
104 connections through 5 shared values. Tap any node.
Inferred from the host name and headers: what a server runs on, not who runs it.
Search 2497 readings taken from 234 arXiv papers' LaTeX source: citation stances (adopts, extends, disputes, measures against) and stated assumptions. Matches on the quoted sentence, the cue phrase that decided it, and the work cited. Returns the sentence verbatim with a link to the exact line.
| Parameter | Type | Description |
|---|---|---|
| limit | integer | Default 15. |
| query * | string | Text to look for, e.g. 'layer normalization' |
| stance | string | Optional filter. The first five are citation stances; bare/cited/shown are what an assumption's own sentence offers for it. |
{
"name": "search_library",
"title": "Search what Keystone read",
"description": "Search 2497 readings taken from 234 arXiv papers' LaTeX source: citation stances (adopts, extends, disputes, measures against) and stated assumptions. Matches on the quoted sentence, the cue phrase that decided it, and the work cited. Returns the sentence verbatim with a link to the exact line.",
"inputSchema": {
"type": "object",
"required": [
"query"
],
"properties": {
"limit": {
"type": "integer",
"maximum": 50,
"minimum": 1,
"description": "Default 15."
},
"query": {
"type": "string",
"description": "Text to look for, e.g. 'layer normalization'"
},
"stance": {
"enum": [
"inherits",
"extends",
"contests",
"compares",
"background",
"bare",
"cited",
"shown"
],
"type": "string",
"description": "Optional filter. The first five are citation stances; bare/cited/shown are what an assumption's own sentence offers for it."
}
},
"additionalProperties": false
}
}None yet. Changes appear after a server's second daily check.
Squares are this server's values; dots are other servers sharing them. A shared address or publisher is a lead, not proof of a common operator.
Everything Keystone reads out of one arXiv paper: what it adopts, extends, disputes and measures against, and the assumptions it states with a citation, its own evidence, or nothing at all. Works for any paper on arXiv that ships LaTeX source — one outside the library is parsed on demand, which takes a few seconds. Nothing is summarised or inferred; every line is the paper's own.
| Parameter | Type | Description |
|---|---|---|
| include | string | Default all. |
| arxiv_id * | string | An arXiv id or link, e.g. 1706.03762 |
{
"name": "read_paper",
"title": "Read one paper's citations and assumptions",
"description": "Everything Keystone reads out of one arXiv paper: what it adopts, extends, disputes and measures against, and the assumptions it states with a citation, its own evidence, or nothing at all. Works for any paper on arXiv that ships LaTeX source — one outside the library is parsed on demand, which takes a few seconds. Nothing is summarised or inferred; every line is the paper's own.",
"inputSchema": {
"type": "object",
"required": [
"arxiv_id"
],
"properties": {
"include": {
"enum": [
"all",
"citations",
"assumptions"
],
"type": "string",
"description": "Default all."
},
"arxiv_id": {
"type": "string",
"description": "An arXiv id or link, e.g. 1706.03762"
}
},
"additionalProperties": false
}
}Every sentence in the library where a paper says prior work is wrong, or puts itself beside it to win. Small on purpose — disagreement is rare and concentrated, and the whole set is readable.
| Parameter | Type | Description |
|---|---|---|
| kind | string | Default disputes. |
| limit | integer | Default 40. |
{
"name": "list_disagreements",
"title": "Where the library argues with itself",
"description": "Every sentence in the library where a paper says prior work is wrong, or puts itself beside it to win. Small on purpose — disagreement is rare and concentrated, and the whole set is readable.",
"inputSchema": {
"type": "object",
"properties": {
"kind": {
"enum": [
"disputes",
"comparisons"
],
"type": "string",
"description": "Default disputes."
},
"limit": {
"type": "integer",
"maximum": 100,
"minimum": 1,
"description": "Default 40."
}
},
"additionalProperties": false
}
}Keystone reads arXiv papers' LaTeX source and reports, verbatim, what each citation adopts / extends / disputes / measures against and what the paper assumes. It is deliberately silent where a sentence carries no cue phrase: an empty result means nothing checkable was said, not that nothing matters. Every reading comes with the sentence and a URL to the line on the page — pass those on, so the person reading your answer can check it.
Names, descriptions and schemas were written by the server and are shown as data.
Shared addresses on big platforms (Cloudflare, Vercel and the like) put unrelated servers together; those are marked shared_infrastructure. A match is a pivot, not a link between operators.