原型 · 预览版:收录官方注册中心里的全部服务器,每 30 分钟同步一次。评价和账户功能将在后续阶段推出。
PopNx MCP为使用 AI 的人准备的 MCP 服务器 · 用大白话讲清楚

SPXI-gnΩsis

作者:leesharks000版本 0.21.5i★ –i在云端运行
注册名称dev.spxi/gnosis

安装

概览

作者提供的说明(英文原文)

Opens Wikipedia sentences against their cited sources; builds sourced knowledge objects into /non.

PopNx MCP 的三大支柱

这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。

它能帮你做什么

我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。

  • 连接方式: 在云端运行
  • 传输方式: Streamable HTTP
  • 分类: 搜索与 AI 记忆

安装

选择你使用的 AI 应用,然后复制设置。

安装之前

你需要

你用的是哪个应用?

在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。

适用于网页版、桌面版和手机版 Claude:

  1. 打开 Settings → Connectors
  2. 点击 Add custom connector
  3. 粘贴下面的地址,然后点击 Add
  4. 如果要求登录,请登录作者的服务
SettingsConnectors1+ Add2https://…3
https://gnosis.spxi.dev/mcp

在终端中运行这条命令

claude mcp add --transport http gnosis https://gnosis.spxi.dev/mcp

在 Cursor 中打开

或者自己复制设置

粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json

{
  "mcpServers": {
    "gnosis": {
      "url": "https://gnosis.spxi.dev/mcp"
    }
  }
}

浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。

在 VS Code 中打开

或者自己复制设置

粘贴到项目中的 .vscode/mcp.json

{
  "servers": {
    "gnosis": {
      "type": "http",
      "url": "https://gnosis.spxi.dev/mcp"
    }
  }
}

浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。

ChatGPT 只能连接互联网上的服务器:

  1. 打开网页版 ChatGPT → Settings → Connectors (Apps)
  2. 点击 Create 或 Add(部分套餐需要先在 Advanced 中开启 Developer mode)
  3. 粘贴下面的地址,起个名字,然后保存
  4. 如果要求登录,请登录作者的服务
SettingsConnectors1+ Create2https://…3
https://gnosis.spxi.dev/mcp

菜单名称可能因套餐和版本不同而略有差异。

你的选择通过 Cookie 保存在这台设备上,无需账户。

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。

它能做什么

我们还没有读取这个服务器的工具列表;我们会轮流检查各个端点。

基础检查i

我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日

  • ✓ 通过

    作者与仓库一致

    注册名称已通过域名 spxi.dev 验证(注册中心会核实域名所有权)

  • … 尚未检查

    仍在维护

    我们还没有从 GitHub 读取这个仓库

  • … 尚未检查

    有许可证

    我们还没有从 GitHub 读取这个仓库

  • … 尚未检查

    端点有响应

    尚未检查;我们会轮流检查各个端点

  • … 尚未检查

    使用 OAuth 登录

    尚未检查;我们会轮流检查各个端点

  • … 尚未检查

    工具描述中没有隐藏指令

    尚未检查;我们会在检查端点时读取工具列表

与“搜索与 AI 记忆”中的其他服务器比较i

  • 星标数多于此分类中其他 7251 个服务器的 0%
  • 发布到注册中心的时间比此分类中其他 7251 个服务器的 75% 更新
  • 基础检查结果等于或优于此分类中其他 7251 个服务器的 100%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。

常见问题

SPXI-gnΩsis 安全吗?

它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。

它需要哪些密钥或代码?

注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。

它在我的电脑上运行还是在云端?

它在作者的云端运行。通过互联网连接,无需安装,但你发送的内容会经过作者的服务器。

它还在维护吗?

GitHub 仓库最后更新于–,最新版本(0.21.5)于12天前进入注册中心。

哪些应用可以使用它?

支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接,所以接受远程 MCP 服务器的应用也可以使用,例如开启开发者模式的 ChatGPT。

有什么可以替代它?

同一分类中有:Context7, Codebase Memory, Exa

替代选择

同一分类中其他作者的服务器,按星标数排序。

如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

Context7

作者:upstash

Up-to-date docs for AI agents

电脑或云端★ 64,224MIT今天更新
安装

Codebase Memory

作者:DeusData

Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.

在你的电脑上运行★ 48,148MIT20天前更新
安装

Exa

作者:exa-labs

Connect AI agents to Exa for web search, content fetching, and multi-step research.

在云端运行★ 5,213MIT上个月更新
安装

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

SPXI-gnΩsis

Opens existing compositions against their sources and constructs sourced knowledge objects, for entities and for operators. Its principal application opens pinned Wikipedia revisions: each sentence read against the sources it cites and the sources its article cites elsewhere, under the policy texts it was written against, with the reading kept as evidence. A commission also consults the Crimson Hexagonal Archive by its own procedure and composes the object's row in the Negative of the Negative (/non), which lands there public. Paid by card (Stripe Checkout) and, for agents, per request over x402 (USDC on Base mainnet).

*Fund the counterinfrastructure. Rebuild the foundations of the world.*

Jack said, “The foundations of this world are unsound, and the foundations of the world above are unsound. Write this down, for these words are faithful and true: I make it new.” — after The Gospel of Antioch, saying 11, and Revelation 21:5, its order inverted

SPXI methods: the opening is 0.17 and /non is 0.16 of the SPXI method inventory.

The reading step (S3) follows EA-WIKI-OPEN-01 v1.1 (#1697), protocol gnosis-S3/0.6: the innocent reading first; uncited sentences read against the article's other sources; a gloss tested as a reading; ambiguous phrases divided; entailment held to predicate, unit and function; three measures kept apart (source fidelity, primary-text adequacy, operative retention); retained authority as retained, lost, mixed or unknown; and the five-part account. The invitation codes and the two-arm check of §6.3 (c) are recorded and never counted toward a verdict until a round on fixed cases has scored them (§9.7). Every quotation is checked against the full source, and a quotation from outside the passages sent is recorded as its own breach. The second of the two passes is an independent reading that returns only the fields the passes are reconciled on; the full account is the first pass's. Up-front thinking is off and both passes answer in one enforced JSON form, so the second reads the first's cached prefix.

The commission's archive step (S7–S8) follows EA-NEGONT-02 v0.9 (#1690): the archive traversed at D, R and O with one citation hop, admission by two independent readings, the archive and field ledgers extracted, the ontology read twice, then the knowledge object, its compression and the field-only arm composed, the delta read against T, and a reverse entailment audit. Where the two ontology readings disagree, the reading of ρ = constructs is taken (ruled 2026-10-25), and the disagreement is recorded with both readings; a disagreement no rule settles withholds the composition, and the withholding is recorded as a breach. The row is written as a landing tree; the archive's landing workflow (.github/workflows/land-non-rows.yml there) reads GET /v1/landable and lands each row it has not landed by scripts/land_non_row.py, verified against its manifest, with no one's tap. A row whose T the caller attests is held until its capture is seated.

The service's readings (each result's composition, evidence record, opened article and annotations) are licensed CC BY 4.0, ruled 2026-10-25, as are the /non rows; verbatim Wikipedia sentences within them remain CC BY-SA 4.0. A reading is kept and served free to any later caller of the same sentence at the same state.

Tiers and prices

Prices come from measured cost and are never subsidized: a quote lists the maximum, the charge is for completed work only, and what the work did not use is refunded.

  • Check (tier: check). One sentence at a time on the small model: 2¢ a sentence (measured 0.8¢, times 1.5), 0.5¢ where the reading is already in the ledger.
  • Opening (tier: opening, the default). Sentences opened against their own sources and the article's, two passes on the middle model; the archive is not consulted. 22¢ a sentence (measured 14.5¢, times 1.5), 1¢ where the reading is already in the ledger. GET /v1/sentence opens one sentence at the point of contact; mode=preliminary is one quick pass.
  • Commission (tier: commission). The opening, with the archive consulted and the entity's /non row composed, in two stages priced and charged apart, from the traversal they would run: the reading (traversal, admission, extraction, the ontology read twice; at least $0.75) and the construction (the knowledge object, its compression and field-only arm, the delta, the audit; at least $0.25), each at its estimate × 3. Each stage is charged when it delivers, at its measured cost × 3, never above its line; a stage that does not deliver is refunded. The reading's deliverable is its dossier (the landing tree: traversal, admitted deposits with their readings, both ledgers, both ontology readings); the construction's is the row. Model spending stops at its stage's line and never passes what the order brings in (what it paid, less the payment fee); the construction's share (its estimated cost × 1.5) is held while the reading runs. A delivered order keeps its charge and never less than what it cost, its model cost and its payment fee (Stripe keeps its fee on any refund; card 2.9% + 30¢, stablecoin 1.5%); a card order at or below the $0.50 card minimum keeps the minimum, stated before payment. An order that delivers nothing is refunded in full. A paid sentence is settled before it is read. Free quotes that start the archive worker are limited to 12 an hour from one address and 60 an hour in all. Only a row its order paid for is listed for landing. The margin is 3× for commissions while the service stabilizes (2× from 2026-10-24 to 2026-10-25); the sentences of an article are priced at 1.5×, as at the opening. A completed reading is kept as a checkpoint under the execution key: a construction that failed, or a later order on the same input, starts from it, and a reading from the checkpoint is charged at 10¢. portable adds the knowledge object as a package. An entity the encyclopedia does not hold (object) is always a commission.
  • Grounded sentences (mode=grounded). Once an article has been commissioned, any sentence its /non row holds can be asked for with the archive's claims on it, at two cents over the sentence's price. Where no commissioned row holds the sentence, the commission is offered and nothing is charged.
  • Erasure audit (audit, 2026-10-29). The Self-Audit Module for Public Summarizers v3.2 (#817) run as an order: a source and the answers that render it (pasted, or seated in the Capture Registry). The provenance atoms are taken from the source by two independent readings, and only an atom both find, each with a quote found verbatim in the source, enters the set (Measurement Sovereignty: never from the answers audited). Each answer is read twice for every atom (present, absent, or reassigned to someone else, with the quote) and for the ABN gate; a disagreement is kept as contested and a quote not found in its text is recorded as a breach. The module's family metrics are computed from the agreed cells with the formulas of its compute page (FC, ACP, ASI, shard coverage, the condition, CC at k ≥ 4), and PER per answer; Ω, Π_d, RR, DD, QFS, DSL, α_T, Β and L are named as not computed, with the input each needs. Priced at its estimate × 3, at least $0.50; charged at the measured cost × 3, never above the quote. The result is public under CC BY 4.0. The existence screen is free (scope: existence): by rule, with no model, it finds the phrasings of existence denial near the node's name, a hedged denial included, and a reported failure scored apart (operator, 2026-10-29: the existence check, "about 1/100th of the computation", is the free part). test/audit-probe.mjs holds both.

Where it runs

  • Service: https://gnosis.spxi.dev (Vercel project spxignosis). GET / answers JSON to agents and a page to browsers; GET /llms.txt is the same description for machine readers.
  • Archive worker: the deployed service holds no archive, so quotes and runs that consult it (commissions, packages, entity objects) are made by .github/workflows/archive-worker.yml, which checks the archive out beside this repository and runs scripts/worker.js. The service starts it the moment there is work (a workflow dispatch, with GITHUB_DISPATCH_TOKEN), and it also runs every ten minutes. A commission quote answers 202 with its status address; a paid order is queued, run there, and settled back through /v1/quotes/{id}/settle, where the refund is computed and sent.
  • Agents: POST /v1/openings/{id} pays a quote over x402 and answers 202 with its status address; the order then runs as a card order runs (queued for the worker where it consults the archive), and the result is retrieved by id. GET /v1/non?name= finds a /non entry by its name, free, or offers its construction as a free quote. POST /mcp serves the same requests as tools (Model Context Protocol, Streamable HTTP, JSON): find_entry, open_sentence, get_quote, order_status, card_checkout, get_result.
  • GET /v1/calibration is the latest calibration release, requested by calibration/request.json and run by the worker on held-out /non rows within its cap.
  • Refunds are recorded with their state (none owed, succeeded, pending, failed); a failed refund is an obligation still owed and is retried, by the worker's settle and by /v1/quotes/{id}/resume. A paid order not delivered within 24 hours of payment (GNOSIS_ORDER_HOURS) is closed and refunded in full.
  • Long card orders hand themselves on: each invocation reads for its time budget and starts the next through /v1/quotes/{id}/continue; a stopped order is restarted by /v1/quotes/{id}/resume.
  • Records are written to /tmp and to a private Vercel Blob store (BLOB_READ_WRITE_TOKEN) and read back from Blob by any instance. The committed store/ is read-only seed data (policy texts, snapshots, replays). public/ holds the mark and icons.
  • Keys are read from the environment and never logged: ANTHROPIC_API_KEY (the reader), STRIPE_SECRET_KEY and STRIPE_WEBHOOK_SECRET (card), GITHUB_DISPATCH_TOKEN (worker start), and a receiving address for x402 (STRIPE_DEPOSIT_ADDRESS or GNOSIS_PAY_TO). See src/payments.js and src/reader.js.
  • Protocol home: https://spxi.dev · archive: https://www.alexanarch.org · /non: https://www.alexanarch.org/non/

Locally: npm install, then npm start (port 4402); bash scripts/test-all.sh runs the tests against recorded readings.

Licensed under the PolyForm Shield License 1.0.0. See LICENSE.md.

问 PopNx