A
alpine-ronin/phpgraph
A knowledge graph of your PHP codebase, for AI coding agents.
★ 6·更新 2026-11-16
在 GitHub 查看 ↗直链 · 无中转页
我们没有从该仓库取到可核对的安装命令,也没有公开的远端端点。这两样都没有时,我们不出配置——编一条出来比不给更糟。 各客户端的接法 →
- ★ 数取值
- 6 ★ · alpine-ronin/phpgraph · 取自 GitHub 搜索 API · (2026-11-19 20:00:48 核对)
- 来源明细
- 收录自 punkpeye/awesome-mcp-servers 榜单(我们从中收集的条目之一) · 条目类型:MCP 服务
- 收录与活跃
- 收录于 2026-11-18 · 最近见到 2026-11-20 · 共见到 3 次 · 被 3 个来源列出
- 评分构成
- 综合分 300 · 6 ★ · 3 个来源 · mcp
常见问题(答案都来自我们的记录)
它适合谁 / 用来做什么?
我们把它归在这些标签下: MCP 服务。标签来自条目自身,不是我们强加的分类。
谁维护它、还在更新吗?
仓库
alpine-ronin/phpgraph;我们收录于 2026-11-18;最近见到 2026-11-20。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。README 原文(英文,未翻译) · 在 GitHub 读完整 README →
A knowledge graph of your PHP codebase, for AI coding agents.
以下摘录自该仓库的 README(片段)
English · Français
A coding agent in a PHP project it does not know searches names with grep, reads file after file, and guesses when
the answer goes through something a name search does not show: an interface, the service container, a message bus,
a route. phpgraph parses the project once and gives the agent the answer from a graph, over MCP, in one call.
In the benchmarks it paid off on the largest project measured, for a question about the impact of a change; on
smaller projects the same kind of question cost more tokens with it.
## Demo
A 68-second video, the same question to Claude Code with and without phpgraph, one real run each:
youtu.be/Cv4ULkN3FF0.
No LLM, no network, no embeddings: your code never leaves your machine, and phpgraph parses it without running it.
Every fact says how sure it is (EXTRACTED, INFERRED or AMBIGUOUS), and what phpgraph could not link is reported.
## Where it helps and where it doesn't
Akeneo PIM (8,416 PHP files), task: list every class that must change when
GetCategoryInterface::byCode() takes a second parameter. Claude Code 2.1.325 with Sonnet, 15 November 2026:
phpgraph 2.3.0, 6 runs; without phpgraph, 3 runs:
| | With phpgraph | Without |
| Correct answers | 6/6 | 3/3 |
| Tokens per run, median | 110,254 | 170,660 |
| Tokens per run, range | 87,643 to 181,273 | 127,193 to 176,783 |
| Time per run | 15.9 to 27.9 s | 15.9 to 26.8 s |
Where it helps: the same answer for 35 % fewer tokens (median) on the largest project; 3 runs out of 6 answered from
impact_of alone, the others checked it with one to three searches. Where it doesn't: on the 6 impact tasks of the benchmark,
with 2.2.0, phpgraph gave more right answers (17/18 against 15/18) but used 9.8 % more tokens in total, from −30 % to
+36 % per task, and time was not better. Results vary from run to run: with 2.2.0 the same Akeneo task took 87,000 to
222,000 tokens.
Every result, grouped by version, with the method: docs/BENCHMARKS.md.
## Quick start
1. Install the executable (Linux, macOS; it carries its own PHP, none needed):
bash
curl -fsSL https://raw.githubusercontent.com/alpine-ronin/phpgraph/main/install.sh | sh
It checks the SHA-256 and installs to ~/.local/bin. Windows, Composer, PHAR and Docker:
docs/INSTALL.md.
2. Connect Claude Code. From your project directory:
bash
phpgraph mcp-config claude
It prints a claude mcp add phpgraph -- ... command with the paths filled in (run it), an .mcp.json to share
with the team instead, and a line for CLAUDE.md telling the agent to call phpgraph before grep. Codex, Cursor,
Sherpa and other MCP clients: phpgraph mcp-config codex|cursor|sherpa|json.
3. Ask. The graph is built on the first call, then kept up to date as you edit.
## What you can ask
What breaks if I change this method? impact_of lists every dependent, nearest first, callers through the
interface it implements included, the routes reaching it and the tests to run.
How does this feature work? outline gives the classes, the routes running into them, the behaviour read in
A coding agent in a PHP project it does not know searches names with grep, reads file after file, and guesses when
the answer goes through something a name search does not show: an interface, the service container, a message bus,
a route. phpgraph parses the project once and gives the agent the answer from a graph, over MCP, in one call.
In the benchmarks it paid off on the largest project measured, for a question about the impact of a change; on
smaller projects the same kind of question cost more tokens with it.
## Demo
A 68-second video, the same question to Claude Code with and without phpgraph, one real run each:
youtu.be/Cv4ULkN3FF0.
No LLM, no network, no embeddings: your code never leaves your machine, and phpgraph parses it without running it.
Every fact says how sure it is (EXTRACTED, INFERRED or AMBIGUOUS), and what phpgraph could not link is reported.
## Where it helps and where it doesn't
Akeneo PIM (8,416 PHP files), task: list every class that must change when
GetCategoryInterface::byCode() takes a second parameter. Claude Code 2.1.325 with Sonnet, 15 November 2026:
phpgraph 2.3.0, 6 runs; without phpgraph, 3 runs:
| | With phpgraph | Without |
| Correct answers | 6/6 | 3/3 |
| Tokens per run, median | 110,254 | 170,660 |
| Tokens per run, range | 87,643 to 181,273 | 127,193 to 176,783 |
| Time per run | 15.9 to 27.9 s | 15.9 to 26.8 s |
Where it helps: the same answer for 35 % fewer tokens (median) on the largest project; 3 runs out of 6 answered from
impact_of alone, the others checked it with one to three searches. Where it doesn't: on the 6 impact tasks of the benchmark,
with 2.2.0, phpgraph gave more right answers (17/18 against 15/18) but used 9.8 % more tokens in total, from −30 % to
+36 % per task, and time was not better. Results vary from run to run: with 2.2.0 the same Akeneo task took 87,000 to
222,000 tokens.
Every result, grouped by version, with the method: docs/BENCHMARKS.md.
## Quick start
1. Install the executable (Linux, macOS; it carries its own PHP, none needed):
bash
curl -fsSL https://raw.githubusercontent.com/alpine-ronin/phpgraph/main/install.sh | sh
It checks the SHA-256 and installs to ~/.local/bin. Windows, Composer, PHAR and Docker:
docs/INSTALL.md.
2. Connect Claude Code. From your project directory:
bash
phpgraph mcp-config claude
It prints a claude mcp add phpgraph -- ... command with the paths filled in (run it), an .mcp.json to share
with the team instead, and a line for CLAUDE.md telling the agent to call phpgraph before grep. Codex, Cursor,
Sherpa and other MCP clients: phpgraph mcp-config codex|cursor|sherpa|json.
3. Ask. The graph is built on the first call, then kept up to date as you edit.
## What you can ask
What breaks if I change this method? impact_of lists every dependent, nearest first, callers through the
interface it implements included, the routes reaching it and the tests to run.
How does this feature work? outline gives the classes, the routes running into them, the behaviour read in
收录徽章(可选)
想把收录标记放进 README 就复制这行,可选,不影响收录。
[](https://anclum.com/zh/s/alpine-ronin-phpgraph/)
- 来源
- awesome · https://github.com/alpine-ronin/phpgraph
- 按领域浏览
- MCP 服务
- Traction
- 6 ★
- 收录来源
- 3 个独立来源
- 收录于
- 今天
更多 MCP 服务器
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。