
Cartographer
io.github.BeppeTemp由 GitHub 账号 BeppeTemp 发布v0.28.0更新于 Nov 7, 2026
Governance MCP server for agent-maintained knowledge bases (OKF): git-backed writes, search, lint.
概览
面向智能体维护知识库的治理型 MCP 服务器:以 git 为底层的共享 wiki,支持检索、校验与写入。
- 可以做什么
- 通过 MCP 工具读取、检索并扩充共享的 Markdown 知识库,带模式校验、链接检查、lint 与每次写入一次提交
- 检查知识库健康状况:失效链接、过期或缺失内容,以及智能体做了哪些改动
- 把知识库中的技能、子智能体、钩子和常驻指令同步到各智能体客户端的原生格式
- 通过插件扩展服务器,添加客户端、命令和 MCP 工具,按提交固定并在批准后启用
- 浏览只读的 Web Atlas,以 3D 图谱查看知识库
- 适用场景
- 适合多个智能体或团队成员把知识沉淀到同一个 wiki,而不是每次会话从零开始;也适合把智能体配置(技能、子智能体、钩子)集中管理并同步到各客户端。安装前可先用官方演示环境试用。
- 运行要求
- 以本地进程运行;清单列出通过 stdio 运行的 Docker 镜像(ghcr.io/beppetemp/cartographer),README 另提供 Homebrew、安装脚本和 Go 构建方式。需要 git 以及一个你拥有的空 git 仓库来存放第一个知识库。未声明环境变量或请求头;团队部署使用 bearer token 角色。
安装
在 SourceWeft 中
- 打开 控制台中的 Cartographer,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
Shared memory for your team's AI agents — one wiki they all read and grow, and none of them can break.
What a colleague's agent works out on Monday, yours already knows on Tuesday — even on another client.
Claude Code, Codex, OpenCode and [any MCP client](docs/mcp-clients.md): one Go binary, git underneath.
[CI] [Release] [License] [MCP] [Listed on mcpservers.org]
Drawn by the Atlas's own graph engine. Pre-1.0: breaking changes bump the minor version and are called out in the changelog.
Try it now
A live Cartographer — no install, no token, no sign-up. It serves two knowledge bases: an astronomy atlas to explore, and a guestbook where visitors' agents leave a message.
1. Connect your agent to https://cartographer-demo.beppetemp.com/mcp
- Claude or ChatGPT — web, desktop or phone: add it as a custom connector, no sign-in (Customize → Connectors → Add custom connector in Claude; Plugins → + → Add custom MCP server, authentication None, in ChatGPT).
- Claude Code:
claude mcp add --transport http cartographer-demo https://cartographer-demo.beppetemp.com/mcp
Codex, OpenCode, Antigravity, GitHub Copilot CLI, Crush, Kiro
2. Ask it something
- "What does this knowledge base know about black holes, and what are they connected to?"
- "Run a health check on the KB: what is broken, stale or missing?"
- "Which skills, subagents and hooks would this KB install in my coding agent?"
3. Leave a message and watch it land
Ask "Leave a message in the guestbook: which client you are and what you would keep in a shared wiki", then open the live Atlas →: your note shows up in the graph next to the ones other people's agents left. Many agents, one wiki.
Everything is read-only except guestbook messages, which are public, anonymous, moderated and rate-limited. To give your agents a knowledge base of their own, see the Quick start.
Why
AI agents start every session from zero, and on a team every colleague's agent starts from its own zero: what one learns stays on that person's machine, in that client. The usual fixes each solve a third of the problem:
- RAG retrieves, but nothing accumulates — the agent never gets to write down what it learned.
- A folder of Markdown the agent edits freely ends in broken links, lost history and silent corruption.
- Skills, subagents and instructions get hand-copied into
.claude/,.codex/,.kiro/… and drift apart on every machine.
Cartographer puts all three in one place: your agents read and grow a knowledge base only through MCP tools, behind a server that enforces the rules, and the same knowledge base carries how your agents are set up.
What you get
-
🧠 Memory that compounds
A wiki the agent builds over time (Karpathy's LLM Wiki pattern): plain Markdown + YAML, one git repo per KB, readable in Obsidian or any editor. No lock-in. -
🛡️ Guardrails, not good intentions
Schema validation, link checks, lint, immutability gates, optimistic concurrency, one commit per write, a signed audit log. The agent cannot leave the KB in a broken state. -
🧩 One KB configures every agent
Skills, subagents, hooks and standing instructions live in the KB and reach each client in its native format (below). Edit once, every machine converges. -
🔌 Extend it with plugins
A plugin is a git repository or a folder that adds clients, skills, subagents, hooks, commands and MCP tools, installed with one command, pinned to a commit and enabled only after you approve what it does. -
👥 Built for teams
Several KBs behind one server — a native user service, a container or a Kubernetes workload — per-KB and per-map access with bearer-token roles, git as the sync layer, conflicts turned into a guided workflow, Ed25519-signed artifacts. -
🗺️ See it all in the Atlas
A read-only web UI in the same binary: a living 3D graph, search, what the agents changed, what the KB ships to them, and what needs fixing. No extra service, no CDN. -
🔑 Secrets, done properly
SOPS-encrypted values referenced by concepts and resolved on demand; plaintext is never stored.
The Atlas, served by the same binary, on a generated demo KB.
Quick start
You need git and an empty git repository you own (GitHub, Gitea, any host) to hold your first KB.
setup shows its plan before touching anything, then installs the local service, creates the KB,
connects every agent client it detects and verifies the result. Restart already-open agent
sessions and ask your agent what it knows. Step by step:
docs/getting-started.md.
Prefer to let your agent install it? Paste this:
If you are an agent asked to install Cartographer: follow that runbook, not the Quick start.
One KB, every agent
cartographer connect detects your installed clients and writes, in each one's own format:
Subagents and hooks are translated, not copied. Every — is a gap declared for a stated reason,
never a silent omission. Cartographer re-syncs at session start (or on a timer where a client has
no hook), restores files edited or deleted by hand, and only ever touches what it created. Exact
paths and reasons: docs/sync.md.
A client missing from the table (a fork, an in-house build) is added by a plugin, without waiting for a release:
see docs/plugins.md.
Install
What lands where, upgrades and removal:
docs/getting-started.md. Server configuration,
environment variables and deployment topologies: docs/deployment.md.
Documentation
Browsable at beppetemp.github.io/cartographer; the
map is docs/index.md.
- Getting started — from zero to a working wiki
- Overview — vision, principles, architecture
- MCP tools — the full tool API
- Deployment — topologies, configuration, backup
- Sync — how artifacts reach each agent client
- Plugins — add clients, commands and MCP tools, approved per version
Contributing
Bugs and ideas go to issues; a
question starts from the FAQ. PRs are welcome: CONTRIBUTING.md
covers building, testing and the PR flow. Cartographer is a personal project maintained on a best-effort basis.
Security reports: SECURITY.md.
License
Apache License 2.0 — see LICENSE.
来源:README.md,提交 2898294
工具
0版本历史
21- v0.28.0最新Nov 7, 2026
- v0.27.3Nov 3, 2026
- v0.27.1Oct 31, 2026
- v0.27.0Oct 29, 2026
- v0.26.1Oct 26, 2026
- v0.26.0Oct 26, 2026
- v0.25.1Oct 20, 2026
- v0.24.0Oct 12, 2026
- v0.23.0Oct 10, 2026
- v0.21.0Oct 7, 2026
- v0.19.2Oct 6, 2026
- v0.19.0Oct 6, 2026
- v0.18.1Oct 4, 2026
- v0.18.0Oct 2, 2026
- v0.17.0Sep 24, 2026
- v0.16.1Sep 23, 2026
- v0.16.0Sep 23, 2026
- v0.15.0Sep 22, 2026
- v0.14.0Sep 18, 2026
- v0.13.0Sep 18, 2026
- v0.12.4Sep 16, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分