20:"$Sreact.view_transition" 21:I[9754,["/_next/static/chunks/2t36f5p4pjzlb.js","/_next/static/chunks/1gyqeidjm5eoa.js","/_next/static/chunks/0_fjnxc4tt_l_.js","/_next/static/chunks/1ymt1shyhmu-4.js","/_next/static/chunks/3kv2zf0i7s2wc.js","/_next/static/chunks/1vs_3g_g75zmk.js","/_next/static/chunks/2g4dh802s8oz3.js","/_next/static/chunks/1bc3f0shukf1j.js","/_next/static/chunks/192kgs9v_t14j.js","/_next/static/chunks/0z6pw0v5c_lx5.js","/_next/static/chunks/08r60e6icmtkc.js","/_next/static/chunks/3trx6c0i7_x-3.js","/_next/static/chunks/40sro5bzp9bi_.js","/_next/static/chunks/1wgf3tk38y6cf.js"],"ProjectDetail"] 22:T4c0d, claude-context — 让整个代码库成为 AI 编码代理的上下文

Zilliz MCP · 代码搜索

让整个代码库
成为编码代理的上下文

Claude Context 是一个为 Claude Code 等 AI 编码代理添加语义代码搜索的 MCP 插件：把数百万行源码在向量数据库中索引，只把相关的那些注入到代理的上下文里。

github.com/zilliztech/claude-context

已注入上下文

relevantModule.search(query)

01 · 机制

整个代码库，收束为一条相关的行

语义搜索从数百万行中直接找出相关代码，省去代理一轮又一轮的摸索，把命中项直接送进 Claude 的上下文。

f"{module.name}: {relevant_snippet}"

02 · 注入约定

索引 · 检索 · 交付进上下文

一条干净的三段流水线：先建立索引，再做语义检索，最后把唯一的相关代码片段直接投放给代理。

INDEX

建立索引

代码存入向量数据库，按语义组织成可检索的向量空间。

RETRIEVE

检索

用查询在向量空间中找到语义最近的相关代码，无需多轮发现。

INJECT

注入

只把相关代码放入上下文，控制大型代码库的上下文成本。

03 · MCP 工具

四个可调用的工具

通过 MCP 对外暴露四组命令，供代理直接调用。

index_codebase

为当前代码库建立语义索引。

search_code

按语义检索代码，返回相关片段。

clear_index

清空已有索引。

get_indexing_status

查询当前索引的状态。

04 · 客户端图谱

一套索引，多种代理

兼容多个 MCP 客户端，包括 Claude Code、OpenAI Codex CLI、Gemini CLI、Qwen Code、Cursor、VS Code 等。

Claude Code OpenAI Codex CLI Gemini CLI Qwen Code Cursor VS Code

05 · 环境与起步

上手需要什么

前置要求

Node.js >= 20
向量数据库，例如 Zilliz Cloud
用于嵌入模型的 OpenAI API 密钥

快速开始

为 Claude Code 配置 MCP，指向 @zilliz/claude-context-mcp
设置环境变量与嵌入模型
配置文件的引入与排除规则

github.com/zilliztech/claude-context/docs

06 · 规模与成本

代码越多，越只取相关

面向大型代码库的 AI 编码场景：存储进向量数据库，仅把相关代码放进上下文，从而让成本可控。

全部源码

语义相关

注入上下文

仅相关代码进入上下文

23:T47e6, claude-context — The entire codebase as context for any coding agent

Zilliz MCP · Code Search

Your entire codebase as the context for any coding agent

Claude Context is an MCP plugin that adds semantic code search to Claude Code and other AI coding agents — indexing millions of lines in a vector database and feeding only the relevant ones straight into the agent's context.

github.com/zilliztech/claude-context

Injected into context

relevantModule.search(query)

01 · Mechanism

An entire codebase, collapsed to the one relevant line

Semantic search finds the relevant code across millions of lines — no multi-round discovery — and carries the hits straight into Claude's context.

f"{module.name}: {relevant_snippet}"

02 · The injection contract

Index, retrieve, deliver into context

A clean three-stage pipeline: index the codebase, retrieve semantically, then deliver the single relevant snippet straight to the agent.

INDEX

Index

Code is stored in a vector database, organized into a searchable semantic space.

RETRIEVE

Retrieve

The query finds the semantically nearest code in that space, without multi-round discovery.

INJECT

Inject

Only the related code enters the context, keeping cost manageable at scale.

03 · The MCP tools

Four tools the agent can call

Exposed over MCP as a small set of commands an agent invokes directly.

index_codebase

Build a semantic index of the current codebase.

search_code

Retrieve code by semantics and return relevant snippets.

clear_index

Clear the existing index.

get_indexing_status

Check the current indexing status.

04 · Client landscape

One index, many agents

Compatible with multiple MCP clients, including Claude Code, OpenAI Codex CLI, Gemini CLI, Qwen Code, Cursor, VS Code, and more.

Claude Code OpenAI Codex CLI Gemini CLI Qwen Code Cursor VS Code

05 · Requirements & getting started

What it takes to get started

Prerequisites

Node.js >= 20
A vector database, e.g. Zilliz Cloud
An OpenAI API key for the embedding model

Getting started

Configure MCP for Claude Code pointing at @zilliz/claude-context-mcp
Set environment variables and your embedding model
Define include & exclude rules for the config

github.com/zilliztech/claude-context/docs

06 · Scale & context-cost optics

More code, less context

Built for AI coding on large codebases: store in a vector database and put only the related code into context, so cost stays in view.

All source

Semantically related

Injected into context

Only related code enters the context

a:[["$","$20",null,{"enter":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-enter","default":"none"},"exit":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-exit","default":"none"},"update":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-update","default":"none"},"default":"none","children":["$","$L21","cmsn44noe004z1s16m0vlzxxq:zh-CN:beautified",{"project":{"id":"cmsn44noe004z1s16m0vlzxxq","slug":"zilliztech-claude-context","name":"claude-context","description":"为 Claude Code 等代理提供语义代码搜索的 MCP 插件","authorName":"zilliztech","authorVerification":{"verified":false,"verifiedAt":null},"category":"development_tools","subcategory":"data_ai_tools","experienceUrl":"https://github.com/zilliztech/claude-context/tree/master/docs","qualityScore":99,"featured":false,"officialSite":null,"views":{"original":{"available":true},"official":{"available":false,"reason":"disabled_by_policy"},"beautified":{"available":true,"reason":null},"defaultView":"beautified"},"beautifiedView":{"revisionId":"cmt2zh47j025562715otj9kvz","kind":"artifact","artifact":{"schemaVersion":"localized-static-html-v4","kind":"static_html","locales":["zh-CN","en"],"defaultLocale":"zh-CN","documents":{"zh-CN":"$22","en":"$23"},"designDirections":{"schemaVersion":"dodot-design-directions-v1","directions":[{"id":"dir-codebase-atlas","name":"Codebase Atlas","projectMetaphor":"The codebase is a surveyable territory — Claude Context turns millions of untracked files into a cartographic atlas of indexed regions that an agent can map and navigate without losing its way.","signatureVisual":"A layered topo/isotype map where each module region is drawn as elevation-type contour rings with a sparse survey legend; a single highlighted 'region' marks the shipped result.","heroGeometry":"A fanning stack of rectangular map-sheets above one overview map composed of per-module contour tiles.","sectionSequence":["Atlas overview: module territories made navigable","Indexing pipeline: how raw files become mapped regions","Search as navigation: a query flying to its region","The MCP tools: index_codebase, search_code, clear_index, get_indexing_status","Client support landscape","Scale, indexing, and cost-conscious navigation"],"surfaceModel":"Stacked map-sheets and paper layers with visible edge lines; contour topography drives the layout, not generic marketing grids.","palette":"Parchment-and-ink survey tones, blueprint contour lines, clay accent for the marked region.","typography":"Survey-technical serif/slab for display, monospace for path and module annotations.","desktopComposition":"Split hero: large topo overview on the left, a stacked-sheet index on the right; region callouts thread through later sections.","mobileTransformation":"Map-sheets become full-bleed tiles with the contour inset; the fan collapses into a single vertical chain of region cards.","mediaStrategy":"No local asset exists in the manifest, so the media region is removed; a derived topo elevation diagram expresses the verify-then-inject relationship, never a fake product screenshot.","forbiddenPatterns":["Browser or editor shell framing","Three-card metric row in the hero","Feature-card wall that any project could fill","Decorative glow or particle field as the signature","Fake tabs, search box, or terminal"],"compositionFingerprint":"codebase-atlas-topo-sheets-v4"},{"id":"dir-retrieval-lens","name":"Retrieval Lens","projectMetaphor":"Claude Context is a convergent retrieval lens: the entire codebase is refracted through semantic search and collapsed down to the single relevant snippet injected straight into the agent's context.","signatureVisual":"A convergence cone of fine source lines that narrow to one focal point — a single highlighted, injected code line — showing millions of lines reduced to one relevant line.","heroGeometry":"A vertical convergence funnel: a wide band of code lines at the top narrowing through the search plane to a single focal line at the bottom.","sectionSequence":["Mechanism: entire codebase converging to one relevant line","The injection contract: index, retrieve, deliver into context","The MCP tools: index_codebase, search_code, clear_index, get_indexing_status","Client support landscape","Requirements and getting started","Scale and context-cost optics"],"surfaceModel":"Lens-system surfaces — overlapping concentric focus rings, a focal-point panel, and a haptic 'injected' result card that materializes the delivered snippet.","palette":"Near-black graphite field with a single warm amber/gold accent used only on the focal injected line; neutral grays in between.","typography":"Monospace for source lines and tool names, condensed grotesque for display headlines.","desktopComposition":"Centered hero: the convergence funnel occupies the focus while an injection-result card to its right shows the delivered line; later sections re-use the focal line as the section spine.","mobileTransformation":"The funnel becomes a vertical stack of descending bars that compress toward a fixed focal line at the bottom; the injection card follows as a focus-stack sheet.","mediaStrategy":"No local asset exists, so the media region is removed; a derived convergence-to-single-line illustration is built from verified flow, never a fake screenshot.","forbiddenPatterns":["Browser, editor, or terminal shell framing","Three-stat metric cards","Generic feature-card wall","Particle, glow, or grid backdrop as the signature visual","Fake search controls or fabricated demos"],"compositionFingerprint":"retrieval-lens-convergence-funnel-v4"},{"id":"dir-neighbor-constellation","name":"Nearest-Neighbor Constellation","projectMetaphor":"The codebase is a constellation of semantic vectors; a query is a new star and its nearest neighbors light up and link, forming the cluster the agent actually reads.","signatureVisual":"A sparse node-field where a central query node connects to its nearest neighbor cluster with explicit links, showing precise vector proximity instead of vague relevance.","heroGeometry":"A radial scatter point-field centered on the query star, with an expanding nearest-neighbor ring cluster around it.","sectionSequence":["Constellation: the codebase as a field of semantic vectors","Embedding and indexing: how the stars are placed","Nearest-neighbor retrieval: the cluster the query lights up","The MCP tools: index_codebase, search_code, clear_index, get_indexing_status","Client support landscape","Requirements and scale of the field"],"surfaceModel":"A deep field of discrete nodes on a void surface, with connection lines drawn only where evidence implies a real neighbor relation.","palette":"Deep indigo near-black void, cool teal nodes, one pure-white query star, sparse cool-links.","typography":"Monospace for nodes and tool names, lightweight grotesque for display.","desktopComposition":"Full-bleed hero point-field with the query star centered; a zoomed retrieval-cluster panel sits below to show which neighbors were read.","mobileTransformation":"The point-field becomes a horizontal scrolling band of stars; the cluster collapses to a single highlighted node with its link label.","mediaStrategy":"No local asset exists, so the media region is removed; a derived nearest-neighbor star diagram is built from verified structure, never a fake screenshot.","forbiddenPatterns":["Editor or terminal shell framing","Three metric cards under a slogan","Feature-card wall","Generic decorative particle field with no relation meaning","Fabricated charts or logos"],"compositionFingerprint":"nearest-neighbor-constellation-cluster-v4"}],"selectedDirectionId":"dir-retrieval-lens","selectionRationale":"The retrieval-lens direction most directly embodies the project's verified mechanism: 'make your entire codebase the context' by converging millions of lines down to the one relevant snippet injected into the agent. Its convergence funnel — wide source lines refracting to a single focal, injected line — maps one-to-one onto the evidence claims of semantic search, direct injection, and context-cost control, whereas the atlas direction emphasizes navigation over the injection outcome and the constellation direction emphasizes vector structure over the delivery moment. The lens also yields the strongest five-second comprehension of the value: many lines collapsing into the one line the agent actually uses. It arrives at a genuinely distinct compositionFingerprint (retrieval-lens-convergence-funnel-v4) not present in the empty antiTemplateContext, and keeps motion purposeful (the converging lines) rather than decorative."},"designBrief":{"schemaVersion":"dodot-design-brief-v1","selectedDirectionId":"dir-retrieval-lens","projectMetaphor":"Claude Context is a convergent retrieval lens: the entire codebase is refracted through semantic search and collapsed down to the single relevant snippet injected straight into the agent's context.","signatureVisual":"A convergence cone of fine source lines that narrow to one focal point — a single highlighted, injected code line — showing millions of lines reduced to one relevant line.","compositionFingerprint":"retrieval-lens-convergence-funnel-v4","heroGeometry":"A vertical convergence funnel: a wide band of code lines at the top narrowing through the search plane to a single focal line at the bottom.","sectionSequence":["Mechanism: entire codebase converging to one relevant line","The injection contract: index, retrieve, deliver into context","The MCP tools: index_codebase, search_code, clear_index, get_indexing_status","Client support landscape","Requirements and getting started","Scale and context-cost optics"],"surfaceModel":"Lens-system surfaces — overlapping concentric focus rings, a focal-point panel, and a haptic 'injected' result card that materializes the delivered snippet.","palette":"Near-black graphite field with a single warm amber/gold accent used only on the focal injected line; neutral grays in between.","typography":"Monospace for source lines and tool names, condensed grotesque for display headlines.","desktopComposition":"Centered hero: the convergence funnel occupies the focus while an injection-result card to its right shows the delivered line; later sections re-use the focal line as the section spine.","mobileTransformation":"The funnel becomes a vertical stack of descending bars that compress toward a fixed focal line at the bottom; the injection card follows as a focus-stack sheet.","mediaStrategy":"No local asset exists, so the media region is removed; a derived convergence-to-single-line illustration is built from verified flow, never a fake screenshot.","evidencePlan":["Render the convergence funnel from the verified index -> retrieve -> inject flow claimed in the summary, keeping every label traceable to material.json.","Present the four MCP tools (index_codebase, search_code, clear_index, get_indexing_status) verbatim from the README headings as factual deliverable commands.","List client support (Claude Code, OpenAI Codex CLI, Gemini CLI, Qwen Code, Cursor, VS Code) and requirements (Node.js >= 20, vector DB, OpenAI API key) exactly as evidenced."],"localeContract":{"zh-CN":"使用中文 slogans、概览、亮点、适用人群与需要注意项，忠于缺省语言文案；六段章节结构保持相同，中文文本密度更紧凑。","en":"Use the English slogan, one-line, overview, highlights, best-for, and caveats verbatim; keep the same six-section structure with natural English density."},"forbiddenPatterns":["Browser, editor, or terminal shell framing","Three-stat metric cards in the hero","Generic feature-card wall that could accept any project's copy","Particle, glow, or grid backdrop serving as the signature visual","Decorative gradients or glows as a substitute for the convergence idea","Fake tabs, search controls, switches, or fabricated demos","Desktop columns merely stacked on mobile without recomposition"],"historicalReuse":{"sourceAssetId":null,"decision":"rebuild","rationale":"material.json contains no historicalSeed, so no isolated historical visual asset exists to source from.","reusedElements":[],"discardedElements":[]}},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"retrieval-lens 方向被忠实实现。签名视觉为收敛漏斗——宽幅源码线经语义搜索收束为一条琥珀色焦点行与注入卡，与 brief 的 signatureVisual、heroGeometry、palette（近黑石墨 + 单一暖琥珀焦点）完全对应。六段章节顺序与 brief sectionSequence 严格一致，四个 MCP 工具逐字、客户端与前置要求均有 material.json 依据。双语文案各自自然且等价，六段结构同步，zh 更紧凑、en 保持原文密度。桌面两栏在 960px/640px 断点处真实重排为单列与堆叠条，未沦为纯堆叠。未回退到通用 Hero/三卡/卡片墙/无依据设备壳/装饰渐变网格——漏斗与焦点行为真正的签名性。","checks":{"projectFit":{"passed":true,"evidence":"六段章节（机制、注入约定、MCP工具、客户端、环境起步、规模成本）严格对应 design-brief.json 的 sectionSequence；hero 复用焦点行作脊线（spine/focal 复用）；历史无素材故为纯净重建，未搬运旧产物。"},"evidenceBound":{"passed":true,"evidence":"MCP 工具四命令（index_codebase、search_code、clear_index、get_indexing_status）逐字来自 README headings；客户端列表（Claude Code、OpenAI Codex CLI、Gemini CLI、Qwen Code、Cursor、VS Code）与 claims 一致；前置要求（Node.js >= 20、Zilliz Cloud、OpenAI API key）与 claims/localizations 一致；包名 @zilliz/claude-context-mcp 出自 README commands；scale 用相对宽度（100/34/6%）仅定性表达收敛关系，未编造统计数字。"},"signatureVisual":{"passed":true,"evidence":"hero 内联 SVG 实现收敛漏斗：顶部宽幅源码线带逐行收窄为单条 #f2b24a 焦点行，配焦点同心环与 inject-card（Injected/已注入上下文 + 着色焦点行），与 brief signatureVisual（数百万行收敛为一条高亮注入线）完全对应，为真实存在且唯一的签名视觉，非截图占位。"},"mobileTransformation":{"passed":true,"evidence":"640px 断点处 flow 3列->1列、tools 2列->1列、reqs 2列->1列、scale 2列->1列，funnel 随宽度缩放，机制段下降条（s1-s5）构成向下压缩聚焦线视觉，符合 brief 的 funnel 变为垂直下降条堆叠、注入卡成焦点堆叠片——非桌面列简单堆叠。"},"localeParity":{"passed":true,"evidence":"zh-CN 与 en 六段结构、视觉与信息完全同步；h1 均用各自 slogan 本地化（中文「让整个代码库成为编码代理的上下文」/ 英文「Your entire codebase as the context for any coding agent」）；zh 文本更紧凑（1211 vs 2500 字节）自然，en 达原文密度，无漏译错译。"},"antiTemplate":{"passed":true,"evidence":"未使用三卡指标行 hero、通用特性卡墙、浏览器/编辑器/终端壳、无依据设备壳、装饰性粒子/发光网格作为签名；背景网格与径向渐变仅为支撑层，签名仍为收敛漏斗；渐变与琥珀发光仅增强焦点行概念，未替代收敛想法。"},"noFakeUi":{"passed":true,"evidence":"repo 与 footer 链接为真实 GitHub 链接；inject-line 为标注「注入上下文」的示意代码而非交互控件；tool 命令为事实性 MCP 命令，其前置 > 仅作命令提示字形、无输入框/无假装可交互的终端；client 为展示式 pill 非假装按钮；qa-report 报 fakeControlCount=0 佐证。"}},"issues":[{"code":"AR-STYLE-CMD_PROMPT","severity":"warning","locale":"all","description":"MCP 工具块的 .cmd::before 渲染一个 > 命令提示字形，虽其下方是真实命令名而非假终端，但略接近终端/伪控件语意。","repairInstruction":"若追求严格，可移除 .cmd::before 的 > 前缀或改为纯文本标签（如 # / +），使 4 个工具更明确是事实命令而非终端提示；非阻塞。"}]},"repairEvidence":[],"provenance":{"executor":"opencode","model":"deepseek/deepseek-v4-flash","opencodeVersion":"1.1.58","promptVersion":"dodot-showcase-deepseek-v4","inputHash":"c77dc64df8b57816d7f88377111fd83eb940bba3178d0df1b9f7cc1e835655da","artifactHash":"661858ddb0b377f244ce461acd9126fe41d70577b444ab4f9cc16817618027e1","designSkill":"dodot-showcase-design"},"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1211,"cjkCount":511,"latinLetterCount":441,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3678,"h1Top":117.015625,"bottomInset":292.15625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"d27741466aee6159ef1fddb9a9f366bcb7c8b71249db779fefc95df5ce3d32b8","middle":"76f67e94e355b3aa146defcdf67cc7b4c4fd0d1fc5ef0abeae9fb89d9ea2630d","bottom":"8b8b180304aa7829b9867e4a267383f28abefeb96df1fce58d69f5741b3e6edd"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1211,"cjkCount":511,"latinLetterCount":441,"scrollWidth":390,"clientWidth":390,"scrollHeight":4686,"h1Top":76.015625,"bottomInset":291.734375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"e799a56acc841d8967e3d9a4012cf30ab35e905b8233c4b6cc641d810d9e1106","middle":"676881b564e10083f95b6408ab5a6b0b3a088b797c00e52ae107f812b0e21b7e","bottom":"22054ee65e4ea8435b3a2ea7e4d27a8d3f9a5c9f76fb19de7d5d96f12447f7a8"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":2500,"cjkCount":0,"latinLetterCount":1992,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3871,"h1Top":117.015625,"bottomInset":270.9375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"40841e0115b32e9e764e2ab5a98493b7d0ea7b76bd02b1729e4241c4e0f094a7","middle":"f43f2d7d8f7464121fc95f0b64ee1c2fccee1a658f7b153925531dcb5c56e37b","bottom":"976cb3c71e991e3f74ff857e7301f9cdb5ef77df0d2c6a86991ba26788dbf8c0"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":2500,"cjkCount":0,"latinLetterCount":1992,"scrollWidth":390,"clientWidth":390,"scrollHeight":5191,"h1Top":76.015625,"bottomInset":291.6875,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"da576716616448f0ba898f4335c8bfc3a2a42ab269f517ced7ec801d48b787e9","middle":"f81b92665f6e713b3005e2ab13d74e78fc3bff6b8a25685854d103073a831d29","bottom":"c1555d188f78f2d4e5371ca55e6a16be04469a83d66a42681b91517c93d0144b"},"consoleErrors":[],"networkRequests":[]}]}},"generatorVersion":"opencode-1.1.58:deepseek/deepseek-v4-flash:dodot-showcase-deepseek-v4","publishedAt":"2026-08-20T11:39:33.310Z"},"rawExtraction":{"projectRevisionId":"cmt2zh47b024z6271852eymz9","projectVersion":1,"sources":[{"canonicalKey":"github:zilliztech/claude-context","role":"primary","revisionVersion":1,"url":"https://github.com/zilliztech/claude-context"}],"facts":[{"key":"repository_owner","value":"zilliztech","evidence":[{"path":"repository.owner","excerpt":"zilliztech"}]},{"key":"repository_name","value":"claude-context","evidence":[{"path":"repository.name","excerpt":"claude-context"}]},{"key":"description","value":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent.","evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]},{"key":"primary_language","value":"TypeScript","evidence":[{"path":"repository.language","excerpt":"TypeScript"}]},{"key":"license","value":"MIT","evidence":[{"path":"repository.license","excerpt":"MIT"}]},{"key":"topics","value":"agent, agentic-rag, ai-coding, claude-code, code-generation, code-search, cursor, embedding, gemini-cli, mcp, merkle-tree, nodejs, openai, rag, semantic-search, typescript, vector-database, vibe-coding, voyage-ai, vscode-extension","evidence":[{"path":"repository.topics","excerpt":"agent, agentic-rag, ai-coding, claude-code, code-generation, code-search, cursor, embedding, gemini-cli, mcp, merkle-tree, nodejs, openai, rag, semantic-search, typescript, vector-database, vibe-coding, voyage-ai, vscode-extension"}]},{"key":"source_count","value":"1","evidence":[{"path":"projectRevision.sources","excerpt":"github:zilliztech/claude-context"}]}],"taxonomy":{"category":"development_tools","subcategory":"data_ai_tools","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]},"lifecycle":{"stage":"active","confidence":0.75,"evidence":[{"path":"repository.pushedAt","excerpt":"2026-07-14T12:00:38.000Z"}]},"userIntents":[{"slug":"build_software","label":"构建或改进软件","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]},{"slug":"automate_work","label":"自动化工作流程","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]},{"slug":"create_content","label":"创作内容或设计","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]},{"slug":"analyze_data","label":"分析与理解数据","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]},{"slug":"learn_explore","label":"学习与探索","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Code search MCP for Claude Code. Make entire codebase the context for any coding agent."}]}],"readme":{"headings":["Your entire codebase as Claude's context","🚀 Demo","Quick Start","Prerequisites","Configure MCP for Claude Code","Configuration","Other MCP Client Configurations","IMPORTANT: the top-level key is `mcp_servers` rather than `mcpServers`.","Optional: override the default 10s startup timeout","**A. Using the Augment Code UI**","**B. Manual Configuration**","Usage in Your Codebase","Environment Variables Configuration","Using Different Embedding Models","File Inclusion & Exclusion Rules","Available Tools","1. `index_codebase`","2. `search_code`","3. `clear_index`","4. `get_indexing_status`"],"commands":["npx @zilliz/claude-context-mcp@latest","npx @zilliz/claude-context-mcp@latest","pnpm install","pnpm build","pnpm dev","pnpm install","pnpm build","pnpm dev","pnpm build","pnpm build:core","pnpm build:vscode","pnpm build:mcp","pnpm benchmark","pnpm dev"]},"changeSet":[{"field":"authorName","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"1074bc10bb9784723844337401bc01cabca6e3b628832b184f5c1fc8c2f41ac9"},{"field":"description","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"3db32c81c208d212d4f5adff51d853a20ff71c83db37a6a915bf05d56ae53b1d"},{"field":"experienceUrl","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"3324ca1b9c8840f3dff8bd02f54d739dba8cc35617a8e904098ba0f836468c3e"},{"field":"name","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"218b2f25aae9678cac9455fc0dd078e4abb63f4c0499891f49121f1b8c45a39c"},{"field":"repository","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"f36ea1c8ec3465ce48b62594e7f074223863e2f753c52f8968372c5f21bbe53d"},{"field":"sourceCount","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"6b86b273ff34fce19d6b804eff5a3f5747ada4eaa22f1d49c01e52ddb7875b4b"}]},"publishedAt":"2026-08-10T10:56:20.052Z","updatedAt":"2026-08-21T13:26:55.396Z","summary":{"id":"cmt2zh47e02516271pjbwcylf","projectType":"product","oneLine":"为 Claude Code 等代理提供语义代码搜索的 MCP 插件","overview":"Claude Context 是一个 MCP 插件，为 Claude Code 和其他 AI 编码代理添加语义代码搜索。它通过向量数据库索引整个代码库，让代理直接获取相关代码片段，无需多轮探索。适用于大型代码库，可降低上下文成本，并支持多种 MCP 客户端。","highlights":["语义代码搜索","支持多种 MCP 客户端","降低大型代码库成本","直接注入相关代码"],"bestFor":["使用 Claude Code 等代理的开发者","大型代码库的 AI 编码场景","希望降低上下文成本的团队"],"caveats":["需要配置向量数据库和 OpenAI API 密钥","需要 Node.js 20+ 环境","依赖外部服务，可能产生费用"],"categorySuggestion":"development_tools","subcategorySuggestion":"mcp_integrations","lifecycleStages":["idea_requirements","frontend","backend_data","testing_quality"],"audienceIntents":["integrate","build","evaluate","discover"],"confidence":0.85,"evidence":[{"claim":"Claude Context 是一个 MCP 插件，为 Claude Code 等 AI 编码代理添加语义代码搜索，将整个代码库作为上下文。","sourceType":"readme","locator":"https://github.com/zilliztech/claude-context"},{"claim":"它使用语义搜索从数百万行代码中查找相关代码，无需多轮发现，直接将结果带入 Claude 的上下文。","sourceType":"readme","locator":"https://github.com/zilliztech/claude-context"},{"claim":"对于大型代码库，它通过将代码存储在向量数据库中，仅将相关代码放入上下文，以控制成本。","sourceType":"readme","locator":"https://github.com/zilliztech/claude-context"},{"claim":"支持多种 MCP 客户端，包括 Claude Code、OpenAI Codex CLI、Gemini CLI、Qwen Code、Cursor、VS Code 等。","sourceType":"readme","locator":"https://github.com/zilliztech/claude-context"},{"claim":"需要 Node.js >= 20，并需要向量数据库（如 Zilliz Cloud）和 OpenAI API 密钥用于嵌入模型。","sourceType":"readme","locator":"https://github.com/zilliztech/claude-context"}],"publishedAt":"2026-08-17T14:18:42.813Z"},"github":{"url":"https://github.com/zilliztech/claude-context","ownerAvatarUrl":"https://avatars.githubusercontent.com/u/18416694?v=4","starsCount":12407,"forksCount":913,"language":"TypeScript","syncedAt":"2026-08-17T14:16:26.981Z","license":"MIT","defaultBranch":"master","archived":false,"disabled":false,"lastChangedAt":"2026-07-14T12:00:38.000Z","missing":false},"tags":[{"name":"Vibe coding","slug":"vibe-coding","verification":"verified","evidenceUrl":"https://github.com/zilliztech/claude-context","evidenceExcerpt":"GitHub topic: vibe-coding"}],"updates":[]},"entryPoint":"ranking","claim":{"state":"unclaimed","verified":false},"initialView":"beautified","community":{"viewer":null,"state":{"likesCount":0,"creativeCount":0,"followersCount":0,"commentsCount":0,"liked":false,"creative":false,"followed":false},"comments":{"comments":[],"nextCursor":null}}}]}],"$L24"]