timesfm
TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
Build, evaluate and serve a bespoke LLM inference engine for your own GPU with crit's commands.
作者提供的原文
教 AI 做这类工作的 Skill,可与本项一起使用
TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
Open-source intelligence for the global theater. Track everything from the corporate/private jets of the wealthy, and spy satellites, to seismic events in one unified interface. Hook an AI agent up to have it parse through data and find previously unseen correlations. The knowledge is available to…
scrape data from Google Maps. Extracts data such as the name, address, phone number, website URL, rating, reviews number, latitude and longitude, reviews,email and more for each place
列表按使用分(满分 100)排序,由以下部分组成
| 热度 | 19/80 | GitHub 星标或每周下载量:较大的计 60,较小的计 20 |
|---|---|---|
| 新鲜度 | 12/12 | 30 天内有发布或代码更新得满分 |
| 仍在响应 | 0/8 | 上次检查时云端端点有响应 |
| 使用分 | 31/100 |
质量: 基础检查通过 3/3 项(仅展示,不用于排序) · 上升势头: 暂无数据 (只用于“上升中”标签)
选择你使用的 AI 应用,然后复制设置。
用一句话让 AI 帮你安装
请帮我安装 crit。按照 https://mcp.popnx.com/i/crit-labs-crit.md 操作,先把命令给我看,等我同意后再运行。粘贴到能在你电脑上工作的 AI(如 Claude Code、Cursor)。它会读取我们的步骤,先给你看命令,等你同意。无需密钥,无需注册。 · 查看 AI 会读取的全文
通过 3/3 项检查 · 在你的电脑上运行 · ✓ 命令已检查 2026年11月20日
第一步:你用哪个 AI?
选择你用的应用,步骤会显示在下方。
可连接 Claude,但需要先准备
需要先在电脑上安装 Python (uv)(只需一次) · 仅限电脑上的 Claude 应用(Windows 或 Mac),网页和手机不可用
把这段话复制到 Claude 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Claude 安装 MCP server crit,按顺序:1) 用 uv --version 检查是否有 uv。没有的话,请让我从 https://docs.astral.sh/uv/getting-started/installation/ 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "mcpServers": { "crit": { "command": "uvx", "args": [ "crit", "mcp" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 crit 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
uv --versionMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: uv 没有点选式安装程序。从 uv 官方页面复制适合你系统(macOS 或 Windows)的一行命令,粘贴到终端 · 不需要单独安装 Python,uv 会在需要时自动下载
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: uv --version
安装一次,所有需要 uv 的工具都能用
SettingsDeveloperEdit Config
{
"mcpServers": {
"crit": {
"command": "uvx",
"args": [
"crit",
"mcp"
]
}
}
}用 crit 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude 说正在使用 crit,或回答中出现 crit 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
ChatGPT 暂时无法连接 crit
crit 需要安装在你的电脑上,而 ChatGPT 只能连接在网上运行的服务器
可连接 Claude Code,但需要先准备
需要先在电脑上安装 Python (uv)(只需一次)
把这段话复制到 Claude Code 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Claude Code 安装 MCP server crit,按顺序:1) 用 uv --version 检查是否有 uv。没有的话,请让我从 https://docs.astral.sh/uv/getting-started/installation/ 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):claude mcp add --scope user crit -- uvx crit mcp 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 crit 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
uv --versionMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: uv 没有点选式安装程序。从 uv 官方页面复制适合你系统(macOS 或 Windows)的一行命令,粘贴到终端 · 不需要单独安装 Python,uv 会在需要时自动下载
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: uv --version
安装一次,所有需要 uv 的工具都能用
Mac:按 Command+空格,输入 Terminal · Windows:打开 PowerShell
claude mcp add --scope user crit -- uvx crit mcp应能看到 crit 在列表中并显示已连接
用 crit 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude Code 说正在使用 crit,或回答中出现 crit 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
可连接 Cursor,但需要先准备
需要先在电脑上安装 Python (uv)(只需一次)
把这段话复制到 Cursor 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Cursor 安装 MCP server crit,按顺序:1) 用 uv --version 检查是否有 uv。没有的话,请让我从 https://docs.astral.sh/uv/getting-started/installation/ 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "mcpServers": { "crit": { "command": "uvx", "args": [ "crit", "mcp" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 crit 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
uv --versionMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: uv 没有点选式安装程序。从 uv 官方页面复制适合你系统(macOS 或 Windows)的一行命令,粘贴到终端 · 不需要单独安装 Python,uv 会在需要时自动下载
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: uv --version
安装一次,所有需要 uv 的工具都能用
用 crit 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Cursor 说正在使用 crit,或回答中出现 crit 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Cursor 帮助页面核实 · 你的界面可能因版本略有不同
可连接 VS Code,但需要先准备
需要先在电脑上安装 Python (uv)(只需一次)
把这段话复制到 VS Code 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 VS Code 安装 MCP server crit,按顺序:1) 用 uv --version 检查是否有 uv。没有的话,请让我从 https://docs.astral.sh/uv/getting-started/installation/ 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "servers": { "crit": { "type": "stdio", "command": "uvx", "args": [ "crit", "mcp" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 crit 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
uv --versionMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: uv 没有点选式安装程序。从 uv 官方页面复制适合你系统(macOS 或 Windows)的一行命令,粘贴到终端 · 不需要单独安装 Python,uv 会在需要时自动下载
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: uv --version
安装一次,所有需要 uv 的工具都能用
用 crit 告诉我你能帮我做什么,然后做一件给我看✓ 如果 VS Code 说正在使用 crit,或回答中出现 crit 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 VS Code 帮助页面核实 · 你的界面可能因版本略有不同
Claude Code
claude mcp add --scope user crit -- uvx crit mcpClaude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"crit": {
"command": "uvx",
"args": [
"crit",
"mcp"
]
}
}
}Cursor (~/.cursor/mcp.json)
{
"mcpServers": {
"crit": {
"command": "uvx",
"args": [
"crit",
"mcp"
]
}
}
}VS Code (mcp.json)
{
"servers": {
"crit": {
"type": "stdio",
"command": "uvx",
"args": [
"crit",
"mcp"
]
}
}
}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。
依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。
在你的电脑上运行,适用于
ClaudeGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
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20日
注册名称已通过 GitHub 账户 crit-labs 验证,该账户拥有此仓库
代码最后更新于昨天
使用 Apache-2.0
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
OSV.dev 中没有 crit@0.17.2 的记录
只检查这个版本的包本身,不包括所有依赖 · 检查于 2026年11月15日
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/servers/crit-labs-crit)提问、分享技巧或报告无法使用。按“有帮助”数量排序。
还没有评价
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。
它在 3 项适用的基础检查中通过了 3 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。
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。
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于昨天,最新版本(0.17.2)于11天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Python (uv)。
同一分类中有:OpenSEO, Exa, Microsoft Learn MCP, Docs, COTAL actions (cotal.ai)
同一分类、其他作者、命令已通过检查。官方的排在前面,再按星标排序。
如何看卡片:↓ 每周下载量i★ 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
Keyword research, site audits, rank tracking and AI visibility for your AI agent.
Official Exa MCP for web search, content fetching, and multi-step research.
Official Microsoft Learn MCP Server – real-time, trusted docs & code samples for AI and LLMs.
cotal.ai actions: product overview, site search, build log, feedback, Cloud waitlist, updates, calls
| 本项 crit | OpenSEO | Exa | Microsoft Learn MCP | Docs | COTAL actions (cotal.ai) | |
|---|---|---|---|---|---|---|
| 运行位置 | 需要安装在电脑上 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 |
| 需要密钥 | 否 | 否 | 否 | 否 | 否 | 否 |
| 可在 Claude 中按 + 添加 | 否 | 是 | 是 | 是 | 否 | 否 |
| 基础检查 | 3/3 | 5/5 | 2/4 | 4/4 | 4/4 | 4/4 |
| 许可证 | Apache-2.0 | MIT | MIT | CC-BY-4.0 | NOASSERTION | Apache-2.0 |
| 价格 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 |
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。
crit_mcp is crit mcp: a Model Context Protocol server that offers crit's command line to an agent as tools, over stdio. An agent whose harness speaks MCP (Claude Code, Codex, Cline and most others) then drives a run through typed tools instead of composing shell commands from AGENTS.md, and waits on the long ones (builds, Eval, references) through jobs the server owns. It needs nothing beyond crit itself.
python -m pip install -e . # from the crit checkout
crit mcp --root /path/to/runs # what the client starts; it speaks MCP on stdin and stdout
Register it with the client, for example Claude Code:
claude mcp add crit -- crit mcp --root /path/to/runs
or a client's JSON configuration:
{"mcpServers": {"crit": {"command": "crit", "args": ["mcp", "--root", "/path/to/runs"]}}}
Run the server where the engines run. A Windows client driving crit in WSL starts it there, with the venv's crit: "command": "wsl.exe", "args": ["-d", "DISTRO", "-e", "/path/to/venv/bin/crit", "mcp", "--root", "/path/to/runs"]. Paths in tool calls are then that machine's paths: give them absolute.
Every subcommand of crit run, crit eval, crit kernels, crit models and crit inference is a tool, and so are crit pull and crit serve: crit run plan is crit_run_plan, crit eval status is crit_eval_status. A tool's input schema is read from the command's own parser when the server starts, so its options, types, choices and help are the command line's and never drift from it. Arguments are the options' long names with underscores (--from-engine is from_engine); a switch is true, a repeatable option a list. Every record option takes the JSON object itself (the server writes it to a file for the command) or a JSON file's path, and a JSON-text option such as quality_budget takes an object. project defaults to the checkout the server runs from (a crit installed without one: the folder the commands run in, --cwd), and root (the run root) of crit run and workspace of crit serve to --root. crit_run_next gives the run's next step and crit_run_show its records.
crit plugins, crit smoke, crit bench and crit hub are not tools: an engine's plugins are files the agent writes, so it runs crit plugins list|add|test and crit smoke --url in its own shell beside the server. crit_run_serve with check runs a candidate's plugin tests and crit smoke as part of the serving check.
| Tool | What it does |
|---|---|
crit_run_* | A run's records: next, create, recon, resolve, plan (check to check it without recording, draft for its skeleton), candidate, build, package, measure, eval, serve, check, seal, select, stop, cost, replay, show, ... |
crit_eval_* | crit Eval: probe, prepare, reference, run, status, kernels, workbench, compare, ab, ... |
crit_pull, crit_serve | Install a crit-hub engine; serve sealed candidates or an installed engine |
crit_kernels_* | List, build and sweep crit's kernel templates on this device; scaffold a workbench for a kernel the engine writes |
crit_models_* | Plan, quantize and verify a weight transform, bind its Eval; distill a drafter (only on the user's word: a run never trains one) |
crit_inference_* | List, search, show and check the Inference techniques |
crit_serve_status | A running crit serve's /health and /v1/models; its api_key goes only to this machine or a private network address |
crit_job_wait, crit_job_output, crit_job_stop, crit_jobs | Follow the commands' jobs |
Every tool carries MCP annotations: its title, readOnlyHint, destructiveHint and openWorldHint, so a client can run the read-only ones without asking and confirm the rest. Read-only tools only read records, logs, files or the device (crit_run_next, crit_run_show, crit_eval_status, crit_kernels_list, crit_inference_*, crit_job_wait, ...). Destructive tools end something or cannot be undone: crit_run_stop, crit_run_seal, crit_run_select, crit_run_recover and crit_job_stop. Every other tool adds records or files and removes nothing. crit_pull, crit_eval_fetch, crit_eval_run and crit_run_eval are open-world: they can download from crit-hub, an engine's git repository and Hugging Face. tools.py holds the table, and a tool it does not name (a newer crit Eval's subcommand) takes the protocol's cautious defaults, destructive and open-world.
Each call runs its command as a job: an owned process tree (a process group, or a Windows job object) whose output and errors go to files. The call waits wait_seconds (default --wait, 30) and then returns the result when the command has ended: its exit code, its JSON output parsed as result (else its output text), the end of its errors (where crit prints reports, such as crit run eval's status and priorities) and the paths of both logs. A command still running returns its job instead, and keeps running: crit_job_wait waits on it again (at most an hour per call, with progress notifications to a client that asks for them), crit_job_output reads either log from any offset and crit_job_stop interrupts it as Ctrl+C would, so crit settles the attempt as interrupted and releases its leases (an interrupted Eval goes on with crit_run_eval and resume), ending the tree if it has not stopped after grace_seconds. On Windows the stop ends the tree at once. Keep wait_seconds below the client's own timeout for a tool call. While Eval or a build holds the GPU, an agent does the next round's CPU work and waits on the job afterwards, as AGENTS.md's long jobs ask.
The jobs belong to the server: when the client closes it (or ends it with SIGTERM) it stops every job still running, so no command outlives the session unseen. A server killed outright cannot: on Linux its jobs then run on, and crit_run_recover proves an abandoned lease.
AGENTS.md and the guides it links (Building engines, Run records, Plugins, Serving, the contribution criteria, Contributing and the CLI, Inference, Kernels, Models, smoke and MCP READMEs) are resources, crit://AGENTS.md, crit://docs/PLUGINS.md and so on, for an agent that does not share the checkout's filesystem. The server's instructions tell the agent to read AGENTS.md first. A crit installed from its wheel carries them all (in crit/library/guides), so crit mcp serves the guides of the release it runs without a checkout.
| Option | Purpose |
|---|---|
--root DIR | Run root for crit run and crit serve calls that name none (default: PROJECT/runs) |
--project DIR | crit checkout the commands use, whose settings.toml gives the device and headroom (default: this checkout; a crit installed without one takes any folder, default the folder the commands run in) |
--cwd DIR | Folder the commands run in (default: the current one) |
--jobs DIR | Folder of the jobs' logs and record files (default: a new temporary folder) |
--wait SECONDS | How long a call waits for its command by default (30) |
--only FAMILY | Offer only these command families' tools (repeatable): run, eval, pull, serve, kernels, models, inference |
--without TOOL | Leave out one tool (repeatable), such as crit_run_cost, which reads the agent's transcript files for their token counts |
The server speaks protocol versions 2024-11-05 to 2025-11-25 over stdio, one JSON message per line. It has no HTTP transport: the agent's harness starts it beside crit.
crit mcp runs crit's own commands on the machine it starts on, as the command line would: they build and run native code (candidates' builds, crit serve, crit Eval's kernels and probes) and keep their records under the run root. The tools reach only these places:
| Where | When |
|---|---|
| The local filesystem | Every command: run records, logs, checkpoints and the folders a call names |
127.0.0.1 | crit serve, crit_run_measure, crit_run_eval and crit_serve_status talk to a local crit serve |
| huggingface.co | crit_pull downloads an engine's checkpoint or prebuilt artifact; crit_eval_fetch, and crit_eval_run or crit_run_eval when the data is missing, download crit Eval's reference package |
| crit-hub.dev and the engine's git repository | crit_pull reads crit-hub's index and the engine's card, and fetches its source at the card's commit, which it builds and runs locally |
crit collects no data and sends nothing to crit-labs: no telemetry, analytics or accounts. The downloads above send the requests any download does (the address and the file asked for) to those hosts, under their own privacy policies. Everything the commands write stays in the folders on your machine that they name, for as long as you keep it. crit_run_cost reads only the token counts and timestamps of the transcript files a call names, on this machine; leave it out with --without crit_run_cost. Questions go to crit's issues.
Prompts an agent with crit mcp handles end to end:
Report problems in crit's issues.