Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Atlas Validate my workflow YAML, dry-run it, and summarize the expected cost."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
中文 · English
Atlas 是一个跑在本机 Windows 上的多模型工作流引擎：用 YAML 画一张有向图，让不同厂商的模型各司其职——并行调研、交叉审查、改代码、人工拍板——每个节点的完整输入输出都实时落在你自己的磁盘上。
它解决一个具体的问题：把多个 LLM 调用串成一条可靠流水线时，"看起来成功了"和"真的成功了"是两回事。Atlas 对每一次调用做假成功检测（空输出、截断、缺字段都过不了关），对每一份产物做哈希断言，对每一笔花销记账。出问题时你能看到是哪一步、哪个模型、为什么。
点开任意节点,可以看到它的完整输入投影、输出产物、实际使用的模型、token 与耗时;运行暂停在人工门时,审批条会直接列出待审材料与完整投影(带哈希),点开即审——批准绑定的就是这份材料的哈希,驳回则必须写明理由。
一张图 = 一个 YAML 文件。比如让三个视角并行分析、汇总后交给编程 agent 改代码、再由人审批：
name: stage-d-custom
nodes:
- { id: left, type: llm, model: Deepseek:deepseek-v4-flash,
prompt: "给出任务的一个侧面要点", consumes: [task] }
- { id: mid, type: llm, model: Deepseek:deepseek-v4-flash,
prompt: "给出另一个侧面要点", consumes: [task] }
- { id: right, type: llm, model: Deepseek:deepseek-v4-flash,
prompt: "给出第三个侧面要点", consumes: [task] }
- { id: joiner, type: llm, model: Deepseek:deepseek-v4-flash,
prompt: "汇总三个要点为执行摘要", consumes: [task, left.output, mid.output, right.output] }
- { id: coder, type: coding_agent, workdir: demo-project,
prompt: "按摘要实施改动并自测", consumes: [task, joiner.output] }
- { id: checker, type: llm, model: Deepseek:deepseek-v4-flash,
prompt: "核对 diff 是否精确完成任务,输出 verdict", consumes: [task, coder.diff] }
- { id: gate, type: human, prompt: "审阅改动与核对结论后批准或驳回" }
edges:
- { from: left, to: joiner }
- { from: mid, to: joiner }
- { from: right, to: joiner }
- { from: joiner, to: coder }
- { from: coder, to: checker }
- { from: checker, to: gate }
- { from: gate, to: END }
guards:
timeout_s: 1800这是一次真实运行的记录（2026-08-19，DeepSeek deepseek-v4-flash）：7 个节点全部完成，编程 agent 精确地在 demo 项目 README 追加了要求的一行文本，diff 审查判定 pass，人工批准后结束。全程 4 分 44 秒，已知成本 $0.0442——账本里每个节点的 token 数、耗时、费用逐条可查。
Related MCP server: Heym
以下均来自发布前用真实供应商 API 跑过的运行，包括失败的（失败也是记录的一部分）。
修复循环 + 代码审查 + 人工门（code-change-review-approve 示例，SuperAI doubao-seed-2.0-lite）。任务："修复 demo-project 中的示例错误并自测"。第一轮：implementer 修好了 fizzbuzz 的判断顺序错误，但顺手多建了两个无关文件，也没真正跑测试；reviewer 输出结构化判定 repair，列出问题清单。第二轮：implementer 清理了多余文件，reviewer 判定 pass。人工批准，结束。两轮共 $0.96，两次 diff 都作为产物留档可下载。（如实说明：当前回边不携带审查意见，第二轮是从冻结 baseline 的重新实施——"循环携带反馈"列在 BACKLOG。）
10 节点自定义图，MCP 直跑（2026-08-22，DeepSeek + SuperAI 四个模型）。对装了 Atlas skill 的 AI 助手描述目标，它当场写了一张 10 节点图：三路并行调研 → 汇总 → 结构化审查（条件路由）→ 修订分支（消费审查意见）→ 复审 → 终稿 → 人工审批 → 门后收尾。整图作为 atlas_run_workflow 的 yaml 参数经 MCP HTTP 端点直接运行，不落盘。8 个执行节点全部一次通过（审查首轮 pass，未进修订轮），人工批准后收尾完成：全程 361 秒（其中 294 秒在等人工），约 $0.01，每个产物的哈希事后独立复验一致。
诚实记录的失败。同一测试矩阵里也出现过：推理型模型把输出预算烧在思考上导致可见文本为空（被空输出检查拦下）、prompt 只送达 1%（被截断哨兵抓到）、agent 首次尝试自报约 $10.5 后自动 retry 被人工终止——这次事故直接催生了现在的成本预留机制和"所有真实 agent 运行必须配预算"的纪律。完整矩阵见 docs/VERIFICATION-2026-08-28.md。
需要 Windows 10/11 x64、Python 3.12、uv。Git clone 用户还需 Node.js 22.12+ 构建一次前端；官方 Release 的 bundle 包已内置构建好的前端（atlas-web 自动识别，也可用 --dist 参数或 ATLAS_WEB_DIST 环境变量显式指定）。
git clone https://github.com/Ctrl1CandV/Atlas.git
cd Atlas
uv sync --locked --all-groups
npm --prefix web ci
npm --prefix web run build
uv run atlas-web一条命令同时启动 Web 界面（http://127.0.0.1:8321）和 MCP 端点（http://127.0.0.1:8321/mcp）。在 Claude Code / ZCode / Cursor 里把 MCP server 指向该 URL，你的 AI 助手就能替你写图、校验、预演、运行；仓库自带的 .mcp.json 也指向该端点。配置细节见 docs/mcp.md。
首次启动会从模板生成本机配置（不覆盖已有文件），凭据只放 config/.env。六个随附示例（并行综合、辩论裁决、map-reduce、重试循环、人工审批管线、代码实施审查）开箱即可校验和预演；真实运行前给每个节点绑定你配置好的模型。
先预演后付费：validate 和 dry_run 零成本，dry-run 与真跑使用同一份有效规格，可用哈希绑定两者身份。
完整性优先：产物按引用传递（文件 + SHA-256），读取时断言哈希；缺失产物显式失败，绝不给下游喂空串；超长不截断而是报错。
假成功即降级：HTTP 200 不算成功。空输出、截断、缺必填字段都会触发跨厂商 fallback 链，降级在界面上显式标注。
全程可审计：append-only JSONL 事件账本是唯一真相，界面显示的一切都能在账本里找到出处；审批证据绑定 baseline/result/patch 三摘要。
崩溃可恢复：控制器被杀后运行自动判定 interrupted，恢复只补未完成节点，成本预留不重算预算。
人在环：human 节点把图暂停在界面里；审批条列出待审材料与完整投影（带哈希，点开即审），驳回必须填写理由。
仅支持 Windows 10/11 x64；以源码 sdist 发布，未上 PyPI，无预编译安装器。当前正式版本 v0.1.0，Git clone 与 sdist 内容略有差异（见 Release 说明）。
Web 只绑回环地址，没有多用户认证；不要暴露到网络。
编程 agent 通过 Claude CLI 以当前用户身份下的宿主进程运行（需 config/agents.json 显式 "runner": "local_cli"，供应商须提供 anthropicBaseUrl 与凭据）。目录副本不是 OS 沙箱：进程理论上能访问当前用户有权访问的任何路径。allowed_paths、回环绑定都不是安全边界。Atlas 不写原目录，diff 由冻结 baseline 的普通文件字节清单生成完整文本 unified diff，二进制变更大声失败。
allow_web: false 只是不授予 Claude CLI 的 WebSearch/WebFetch 工具；可写 agent 有 Bash，仍可能联网。max_turns 是校验过的规格字段，但当前 Claude CLI 没有硬轮次参数，硬限制来自 deadline 与已配置预算。
research/coding_agent 节点缺省不自动重跑（retry 缺省 0）；显式声明 retry: N 后，dry-run 必须出现放大风险警告。
成本帽只在费率已知的调用上精确生效；费率未知时保守占满剩余预算，但不能证明供应商实际账单没超。
"多厂商辩论"等名称只表达拓扑；只有绑定了真实不同的供应商，意见才独立。
审批证据绑定 baseline_digest、result_digest 与 patch_digest 三摘要；可写 coding agent 与 allowed_paths 的组合会在创建 run 前被拒绝（Claude --add-dir 不是只读边界）。
2026-08-28 基线：Python 测试 666 passed / 2 skipped（另有 5 个真实供应商测试默认排除、需主动运行且可能收费）；Web 测试 22 passed、lint 0 告警、生产构建通过；公开 CI 在 main 分支三 job 通过（另有 workflow_dispatch/release 触发的可选浏览器 e2e 冒烟 5 用例——键盘审批流跑 Chromium/Edge/Firefox 三浏览器 + 终局卡片 dark/light × 100%/200% 四组合基线截图，不计入主门）。10 节点自定义图经 MCP HTTP 端点真实运行全链路通过（含人工审批，见上文真实案例）。
2026-08-19 发布基线：Python 测试 427 passed；Web 测试 22 passed、lint 0 告警、生产构建通过；六个示例工作流严格离线 validate/dry-run，0 次供应商调用；发布 sdist 100 个条目、0 扫描发现。数字对应当时的源状态；当前基线与真实端到端证据见 docs/VERIFICATION-2026-08-28.md。
开发验证：
uv lock --check
uv run python -m compileall -q atlas
uv run pytest
npm --prefix web test && npm --prefix web run lint && npm --prefix web run builddocs/VERIFICATION-2026-08-28.md — 验证报告：基线、真实端到端证据、已知发现
docs/mcp.md — MCP 工具面与调用口径
docs/mcp.md — harness 接入配置
skill/SKILL.md — 给 AI 助手看的操作手册
请求取消一个运行(P2,零成本)。
done/failed/cancelled 返回冲突;running 写请求后 controller 在 下一节点边界终止(在途模型调用只能等它返回或超时,不宣称强杀); paused/interrupted 在锁内直接落 run_cancelled 终态。幂等。
| Name | Required | Description | Default |
|---|---|---|---|
| reason | No | ||
| run_id | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and does so well: it declares the operation idempotent, zero-cost (P2), discloses that in-flight model calls are not force-killed but waited out or timed out, and explains the lock-based immediate finalization for paused/interrupted runs. This is unusually rich operational disclosure.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The purpose is front-loaded in the first clause, followed by compact state-machine details. Dense but every sentence adds behavioral information; no filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists, so return values need not be described, and the state-transition semantics are thoroughly covered. The only real gap is parameter documentation, especially the unexplained 'reason' field.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0% and the description never mentions either parameter. run_id is inferable from the name, but the 'reason' parameter is completely undocumented in both the schema and the description, leaving the agent no guidance on its format or purpose.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The opening clause states a specific verb and resource ('请求取消一个运行' / request cancellation of a run), which clearly distinguishes it from siblings like atlas_resume_run and atlas_run_workflow. The scope is unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives state-conditioned guidance: terminal states (done/failed/cancelled) return conflict, running defers to the next node boundary, and paused/interrupted finalize immediately. This tells the agent when the call is meaningful, but it never names alternatives such as resume or get_run for related tasks.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
查某次运行的动态状态、每节点模型与 token、账本路径(零成本)。
| Name | Required | Description | Default |
|---|---|---|---|
| run_id | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden. It does disclose a useful behavioral trait beyond the schema: the operation is zero-cost (零成本). However, it says nothing about read-only nature, auth requirements, or rate limits, so behavioral coverage remains partial.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single tight sentence, front-loaded with the verb and followed by the returned fields and the cost note. No wasted text, though it is quite terse.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists, so return-value explanation is not strictly required, and the tool is a simple single-parameter getter. Missing usage context and parameter detail against zero annotation coverage leave it only minimally adequate.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0% and the sole parameter run_id has no documentation in either schema or description. The phrase '某次运行' only implicitly maps to run_id and gives no format or provenance (e.g., where to obtain the id), so the description does not compensate for the coverage gap.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb (查) and resource (某次运行的动态状态) and enumerates the returned data: dynamic status, per-node model and tokens, and ledger path. An agent can distinguish it from the list-style siblings, though it does not explicitly name an alternative. This is clear but not sibling-differentiated in text.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
There is no guidance on when to use this tool versus alternatives such as atlas_list_runs or atlas_cancel_run. Usage must be inferred from '查某次运行'. No prerequisites or exclusions are stated.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
列出运行(零成本,P4)。按 run_id 降序稳定分页。
每条含 run_id/graph/status/nodes_done/started;状态为 running/interrupted/paused/done/failed/cancelled(interrupted 由 账本+运行锁动态判定)。starting 只在账本落账前的短暂窗口由 Web 单 run 查询 合成,不出现在本列表。cursor 传上一页返回的 next_cursor 续页。
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No | ||
| cursor | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and adds substantial behavioral context: zero cost, P4, stable pagination, the full status enumeration, dynamic determination of 'interrupted' via ledger and run lock, and the short-lived 'starting' window excluded from this list. It does not cover authentication or rate limits, but the disclosure is rich for a read-only list operation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is front-loaded with purpose and ordering, then adds status details and pagination rules. Every sentence serves a purpose for a tool with zero annotations, but some status enumeration could be relegated to the output schema since one exists.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists, so return fields needn't be explained, yet the description still lists key fields. It covers pagination, sorting, statuses, and the subtle 'starting' exclusion. Gaps remain around the 'limit' parameter and authentication context, but overall it is fairly complete for a list tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description must compensate. It explains 'cursor' (pass the previous page's next_cursor), but completely omits the 'limit' parameter—its default (20 in schema) and any bounds or semantics. Half the parameters remain undocumented.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
State a clear verb+resource: '列出运行' (list runs), with sorting by run_id descending and stable pagination. It enumerates returned fields (run_id/graph/status/nodes_done/started), making the purpose unambiguous. However, it does not explicitly distinguish from sibling tools like atlas_get_run.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives pagination mechanics ('cursor 传上一页返回的 next_cursor 续页') but no guidance on when to use this tool versus alternatives (e.g., atlas_get_run for a single run, or atlas_list_workflows for workflows). No when-not conditions are stated. Usage is only implied by the tool name.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
列出 workflows/ 里的图定义与校验状态(零成本)。
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations, so the description carries the burden. It discloses two useful behavioral traits: it is zero-cost and it inspects validation state. However, it doesn't disclose read-only nature explicitly, pagination, or the format of the returned list.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single compact sentence, front-loading the action and constraint. No filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists, so return values need not be explained. For a zero-param read, the description plus output schema covers the essentials, though the lack of sibling disambiguation is a minor gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Zero parameters, so the baseline is 4. There is nothing to document and the description correctly implies a no-argument call.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb+resource (list workflow graph definitions) and adds scope (in workflows/ directory) plus what's returned (validation status). Clear enough, though it doesn't explicitly contrast against sibling atlas_validate_workflow or atlas_list_runs.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The '(zero-cost)' hint implies safe/cheap enumeration, hinting at when to use it, but there is no explicit when-to-use vs alternatives guidance such as preferring this over atlas_validate_workflow for a quick status check.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
仅恢复动态判定为 interrupted 的运行；paused 必须在 Web 审批。
| Name | Required | Description | Default |
|---|---|---|---|
| run_id | Yes |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full behavioral burden. It discloses a non-obvious state gate (interrupted vs paused), which is genuinely useful, but says nothing about permissions, what happens if the run_id is not in an interrupted state, or reversibility of the resume.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single short sentence with the precondition front-loaded and zero filler. Nothing could be removed without losing information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists, so return values need not be described, and the key precondition is covered. Still, for a state-mutating tool with no annotations, the description leaves error behavior and permission requirements unaddressed.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0% and the description never mentions run_id, so neither source explains the parameter's format or origin. Compensation for the coverage gap is absent, though the parameter's meaning is close to self-evident from its name.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource in Chinese: resume a run. The scope qualifier ('only runs dynamically judged as interrupted') sharpens the action beyond the bare name. It does not, however, distinguish itself from siblings like atlas_cancel_run or atlas_get_run by name or contrast.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly states the condition for use (only interrupted runs) and the when-not case (paused runs must go through Web approval). This routes the agent to the correct path before it ever calls the tool, which is exactly what usage guidance should do.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
运行工作流。同步阻塞到结束。
workflow_id 与 yaml 二选一(都传时以 yaml 为准):
workflow_id:运行已保存的 workflows/.yaml;
yaml:运行即时编写的自定义图全文,不落盘、不写 workflows/。 persist_as(可选):真实运行结束后,把 yaml 固化为 workflows/.yaml——新建要求 id 未被占用,更新需 expected_sha256 语义与 atlas_save_workflow 相同;dry_run 不固化。
attachments(可选,E-2A):运行附件 [{name, path}];path 是发起 机器本机绝对路径,启动前整读校验(单件 ≤16MiB/合计 ≤32MiB,名字 全小写且不得撞保留名/节点 id),内容字节克隆进 run 产物库后与原 文件再无关联。下游节点用 consumes 引用 name;投影只含摘要行, 原字节经界面产物工作台查看。响应不含原始路径。
node_overrides 是封闭的本次运行节点参数覆盖；不改 YAML，不接受权限或拓扑字段。 可覆盖:model/fallback/thinking/max_output_tokens/temperature/seed/ timeout_s/retry/prompt(llm)、max_turns/timeout_s/retry/prompt/workdir (coding_agent;research 无 workdir)、prompt(human)。 prompt 是完整替换本次运行的节点职责文本,不是追加; consumes/outputs/图结构永远不可覆盖——要长期生效就用 persist_as 固化 或调用 atlas_save_workflow。 dry_run=True 与真跑使用同一有效规格，只渲染、不花钱； 默认同步阻塞到结束(wait=true);wait=false 见下。dry_run=False 真实调用模型。暂停在 human 节点时返回 paused。 wait(默认 true)同步阻塞到结束;wait=false 通过全部预检与执行身份 断言后立即返回 run_id(status=starting),长任务不再占住会话—— 用 atlas_get_run 轮询、atlas_list_runs 列表。wait=false 与 persist_as 互斥(固化需要等真跑结束)。
| Name | Required | Description | Default |
|---|---|---|---|
| task | Yes | ||
| wait | No | ||
| yaml | No | ||
| dry_run | No | ||
| persist_as | No | ||
| attachments | No | ||
| workflow_id | No | ||
| node_overrides | No | ||
| expected_execution_sha256 | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and does so: it discloses cost behavior (dry_run=False actually calls models), blocking semantics, pause-at-human return value, attachment byte-cloning with size/name constraints, that raw attachment paths are excluded from the response, and the status=starting return for wait=false. This is unusually rich behavioral context.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The purpose and blocking behavior are front-loaded, and the segment structure (one cluster per parameter) makes a long description scannable. The density is largely justified by 9 parameters, though some clauses could be tightened without loss.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists so return values need not be spelled out, and the description still adds useful response notes (no raw paths, paused status). Combined with the parameter and cost disclosures it is nearly complete; the unexplained required 'task' parameter is the one real hole for a tool of this complexity.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0% and there are 9 parameters, so the description must compensate; it explains workflow_id, yaml, persist_as, attachments, node_overrides, dry_run, wait and references expected_sha256 semantics. However, the single required parameter 'task' is never mentioned or defined anywhere, leaving its meaning and expected content unresolved.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States the specific verb+resource (run a workflow) and immediately qualifies the scope: synchronous blocking to completion, workflow_id vs inline yaml, and dry_run vs real execution. An agent can distinguish it from atlas_validate_workflow, atlas_save_workflow and atlas_resume_run without opening any schema.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly routes the agent: workflow_id vs yaml (yaml wins if both), persist_as for long-term effect vs atlas_save_workflow, dry_run for rendering without cost, and wait=false for long jobs polled via atlas_get_run/atlas_list_runs. It also names the exclusion (wait=false is mutually exclusive with persist_as).
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
把校验通过的 YAML 保存为 workflows/.yaml(零成本)。
校验不过不保存。新建要求 id 未被占用;更新必须传 expected_sha256 (上次读到的文件哈希)——文件被改过就拒绝,防静默覆盖。 返回 file_sha256(下次更新用它)与 spec_fingerprint。
| Name | Required | Description | Default |
|---|---|---|---|
| yaml | Yes | ||
| workflow_id | Yes | ||
| expected_sha256 | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden and does so well: it discloses that invalid YAML is not saved, that concurrent modification is rejected via expected_sha256 (preventing silent overwrite), and that it is zero-cost. It omits any permission/auth requirements, which keeps it short of a 5.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Front-loads the core action and path, then layers the validation gate, create/update conditions, and return values in short, load-bearing clauses. Minimal waste; the parenthetical '零成本' earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists, yet the description still adds return semantics (file_sha256 reused for next update, spec_fingerprint), which is genuinely useful. Failure modes and concurrency behavior are covered; only auth/permission context is missing.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, so the description must compensate. It explains expected_sha256 thoroughly (last-read file hash, optional for create) and implies workflow_id via the file path, but the yaml parameter's content/format is never described.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb (save), the exact resource and destination path (workflows/<workflow_id>.yaml), and the precondition that only validated YAML is persisted. An agent can distinguish this from siblings like atlas_validate_workflow or atlas_run_workflow immediately.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Clearly separates the two usage modes: new creation requires an unoccupied id, while update requires passing expected_sha256. It does not explicitly name atlas_validate_workflow as the prerequisite step, but the '校验通过的 YAML' framing makes the workflow order clear.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
校验一份图定义(零成本)。传 yaml 全文,或已保存的 workflow_id。
检查:YAML 语法、节点类型封闭清单、边引用、条件边与路由字段、入口、 可达性、死环、有环必须设 max_iterations、consumes 引用、异质性提示。 返回 JSON；校验不过时 error 会指出具体字段或图结构问题，并统一附带 YAML path、line、column（整图聚合错误没有唯一坐标时只返回 path）。 按 workflow_id 校验时额外回显 yaml 原文与 file_sha256——修改已有图的 读-改-写闭环:改 yaml 内容后带 file_sha256 作为 expected_sha256 调 atlas_save_workflow。两个参数都传时以 yaml 全文为准。
| Name | Required | Description | Default |
|---|---|---|---|
| yaml | No | ||
| workflow_id | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations present, the description carries the full burden and does so: it discloses zero cost, the full validation checklist, the error format including YAML path/line/column (and the path-only fallback for aggregate errors), and the extra echoed fields (yaml text, file_sha256) when validating by workflow_id. This is far beyond what the schema conveys.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The purpose and cost note are front-loaded, and every subsequent sentence carries information (checks, error format, echo fields, precedence). It is dense but not padded; the long enumeration of checks is justified for a validator.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
An output schema exists so return values need not be explained, yet the description usefully adds error-format and echo-field detail. For a two-parameter validator in a sibling set that includes save/run/get, the description covers everything needed to invoke it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description must compensate, and it does: it explains the meaning of both empty-defaulted parameters, their two usage modes, the precedence rule when both are supplied, and the additional return content tied to workflow_id. That is meaningful semantic detail the bare schema lacks.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource ('校验一份图定义') and then enumerates the exact checks performed (YAML syntax, node type closed list, edge references, conditional edges, entry, reachability, dead loops, max_iterations, consumes references). This distinguishes it clearly from siblings like atlas_save_workflow or atlas_run_workflow.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly gives the two invocation modes (full yaml vs. saved workflow_id) and resolves the overlap by stating yaml text takes precedence when both are passed. It also names the correct downstream sibling and the exact handoff (pass file_sha256 as expected_sha256 to atlas_save_workflow), which is precisely the routing an agent needs.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.2.0atlas_cancel_runatlas_get_runatlas_list_runsatlas_list_workflowsatlas_resume_runatlas_run_workflowatlas_save_workflowatlas_validate_workflowScored across 8 tools
Each tool has a clear, distinct purpose in the workflow lifecycle: get run status, resume interrupted runs, validate/save workflows, run/list/cancel runs, list workflows. No two tools are easily confused. The only slight overlap is between atlas_run_workflow and atlas_save_workflow regarding persistence semantics, but they are clearly differentiated.
All tool names follow a consistent 'atlas_verb_noun' pattern (get_run, resume_run, validate_workflow, save_workflow, run_workflow, list_runs, list_workflows, cancel_run). This is highly predictable and readable.
8 tools is well-scoped for a workflow orchestration server. Each tool fills a specific role in the workflow lifecycle (validation, persistence, execution, monitoring, cancellation), and there are no redundant tools.
The surface covers the full workflow lifecycle: validation, saving, running, listing runs/workflows, getting run details, resuming, and canceling. One notable gap is the absence of a delete/remove operation for workflows, which would be expected for full lifecycle management. However, core operations are complete and agents can work around this.
Open-source Zapier/n8n alternative as an MCP server: agents build, run and debug your workflows.
LLM Orchestration Agent (Mcp)
Control plane for autonomous software labor. Agents claim objectives over MCP with audit trail.
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.
