[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-80685":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":15,"stars7d":16,"stars30d":16,"stars90d":14,"forks30d":14,"starsTrendScore":17,"compositeScore":18,"rankGlobal":10,"rankLanguage":10,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":22,"hasPages":20,"topics":23,"createdAt":10,"pushedAt":10,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":14,"starSnapshotCount":14,"syncStatus":29,"lastSyncTime":30,"discoverSource":31},80685,"pgr","entireio\u002Fpgr","entireio","PGR is an experimental, stateless MCP code-search server for studying how ranking, latency, and output shaping affect agentic search.","",null,"Python",53,1,0,4,6,12,52.5,"MIT License",false,"main",true,[24,25],"mcp-server","research","2026-06-12 04:01:29","![pgr repository cover](assets\u002Fgh-repo-cover.png)\n\n# pgr\n\n`pgr` is an experimental, stateless MCP server for code search, built as part of\nresearch into how coding agents search repositories.\n\nIt is a working prototype and public research artifact, not a polished product.\n\nIt wraps local code search with ranking and output shaping designed for agent\nworkflows: surfacing likely implementation files earlier, de-prioritizing tests\nand low-value matches, and formatting results so a model can decide what to read\nnext with less thrashing.\n\nThis repository also includes the public datasets, benchmark packages, and saved\nresults used in the accompanying writeup on agentic code search.\n\n- [Quick Start](#quick-start)\n- [Usage Documentation](#usage-documentation)\n- [Public Benchmarks and Data](#public-benchmarks-and-data)\n- [Contributing guide](CONTRIBUTING.md)\n- [Security policy](SECURITY.md)\n- [License](LICENSE)\n\n## Quick Start\n\n### 1. Install pgr\n\nInstall `pgr` from this repository:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fentireio\u002Fpgr.git\ncd pgr\ncargo install --path .\n```\n\n### 2. Add it to your MCP client\n\nThen add it to any MCP client that can launch a stdio server:\n\n```json\n{\n  \"mcpServers\": {\n    \"pgr\": {\n      \"command\": \"pgr\",\n      \"cwd\": \"\u002Fabsolute\u002Fpath\u002Fto\u002Frepo\"\n    }\n  }\n}\n```\n\n### 3. Point it at the repo to search\n\nSet `cwd` to the repository you want your agent to search. `pgr` searches the\ncurrent working directory of the process that launches it.\n\n### 4. Ask your agent to search\n\nOnce configured, ask your agent to search the repo. For example:\n\n```text\nsearch for where checkpoint commits are written\n```\n\n```text\nfind the code that handles MCP tool calls\n```\n\nIf `pgr` is not yet on your shell `PATH` after install, use the binary directly\nat:\n\n```bash\n~\u002F.cargo\u002Fbin\u002Fpgr\n```\n\n## What pgr gives agents\n\n`pgr` is not trying to replace `ripgrep`. It shells out to local `rg` for fast\ncontent search and file listing, then ranks and shapes the results for coding\nagents.\n\nThat means an agent is more likely to see implementation files before test\nnoise, repeated references, or lower-signal matches.\n\nAt a high level it provides tools for:\n\n- `search_code`\n- `read_code`\n- `find_files`\n- `list_dir`\n\nThe search output profile is controlled with `PGR_OUTPUT_PROFILE`. If it is\nunset, the server defaults to the richer planner-oriented output profile used in\nthe public results.\n\nThere is no index to build, no background daemon, and no separate storage layer.\n`pgr` starts, answers MCP tool calls over stdio, and exits when the client stops\nit.\n\n## Build from source\n\n```bash\ncargo build --release\n```\n\nThe binary will be available at `target\u002Frelease\u002Fpgr`.\n\nIf you want the binary on your `PATH`, you can also install it locally with:\n\n```bash\ncargo install --path .\n```\n\nFor most first-time users, this is the easiest way to try `pgr`.\n\n## Requirements\n\n- Rust toolchain for building from source\n- [`ripgrep`](https:\u002F\u002Fgithub.com\u002FBurntSushi\u002Fripgrep) installed and available as\n  `rg`\n\n`pgr` shells out to local `rg` for both content search and file listing. If `rg`\nis not installed, `pgr` will not be able to search.\n\n## Why we built it\n\n`pgr` is built around a simple idea: faster search alone is not enough. What\nmatters just as much is whether the search result helps the model choose a\nbetter next action.\n\nIn the public benchmarks in this repository, the strongest effects did not come\nfrom raw scan speed. They showed up in more local agent behavior:\n\n- better first-query retrieval quality\n- fewer redundant search loops before the first code read\n- getting the agent into reading relevant files sooner\n\nThat is the problem `pgr` is designed to improve.\n\n## Usage Documentation\n\n- [docs\u002Fusage.md](docs\u002Fusage.md) - MCP setup, direct stdio smoke test, tool\n  arguments, and output profiles\n\n## Public benchmarks and data\n\nThe public artifacts referenced in the writeup live under\n[`public_release\u002F`](public_release\u002FREADME.md).\n\nKey packages:\n\n- [`public_release\u002Fdata\u002Fentireio_cli_checkpoints_2026_04_15\u002FREADME.md`](public_release\u002Fdata\u002Fentireio_cli_checkpoints_2026_04_15\u002FREADME.md)\n  - public `entireio\u002Fcli` checkpoint export used for the trace analysis\n- [`public_release\u002Fbenchmarks\u002Fentireio_cli_fff_vs_baseline_public60\u002FREADME.md`](public_release\u002Fbenchmarks\u002Fentireio_cli_fff_vs_baseline_public60\u002FREADME.md)\n  - speed-oriented benchmark comparing `ripgrep` and `fff`\n- [`public_release\u002Fbenchmarks\u002Fentireio_cli_ranking_public60\u002FREADME.md`](public_release\u002Fbenchmarks\u002Fentireio_cli_ranking_public60\u002FREADME.md)\n  - broad mixed-workload benchmark comparing `baseline`, `fff`, and `pgr`\n- [`public_release\u002Fbenchmarks\u002Fentireio_cli_offline_ir_public60\u002FREADME.md`](public_release\u002Fbenchmarks\u002Fentireio_cli_offline_ir_public60\u002FREADME.md)\n  - offline retrieval replay benchmark built from real agent-issued queries\n\n## What is in this repo\n\n- `src\u002F`\n  - the Rust MCP server\n- `tests\u002F`\n  - integration tests for the MCP surface\n- `eval\u002Fv2\u002F`\n  - the minimal evaluation harness and backend adapters used by the public\n    benchmark runners\n- `public_release\u002F`\n  - public datasets, benchmark definitions, summaries, and saved outputs\n    referenced in the blog post\n\n## Repository scope\n\nThis public repository is intentionally focused on:\n\n- the Rust MCP search server\n- the minimal benchmark harness needed by the public runners\n- the public data and results packages cited in the writeup\n\nOlder internal drafts, duplicate exports, and superseded experimental outputs\nhave been removed so the repo stays small, inspectable, and externally\nunderstandable.\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md), [SECURITY.md](SECURITY.md), and [LICENSE](LICENSE).\n","pgr 是一个实验性的无状态MCP代码搜索服务器，旨在研究排名、延迟和输出格式如何影响代理搜索。该项目使用Python编写，通过与本地代码搜索工具（如ripgrep）结合，为编码代理提供优化的搜索结果排序和格式化处理，优先展示可能的实现文件，并降低测试文件和低价值匹配项的优先级。pgr适用于需要高效代码搜索的研究场景或开发环境，特别是那些关注于提高自动化代码理解与操作效率的应用。此外，它还提供了公共数据集和基准测试包以支持相关研究。",2,"2026-06-11 04:01:39","CREATED_QUERY"]