[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74PRTSMP5KNNCHP93N3G50T":3,"skill-same:01M74PRTSMP5KNNCHP93N3G50T":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M74PRTSMP5KNNCHP93N3G50T","create-mcp-server",null,"Build, iterate, and evaluate Model Context Protocol (MCP) servers that expose external services as tools an LLM can call. Use when asked to \"build an MCP server\", \"create an MCP tool\", \"wrap this API as MCP\", \"expose X to Claude\", or when extending OTTO with custom tool integrations. Covers research, schema\u002Ftool design, error handling, pagination, testing via MCP Inspector, and producing a 10-question eval set that proves the server actually enables real work.","cmetech\u002Fotto-cli","cmetech","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F118027304?v=4","01M74PNNHCP8T9GH9NWAWHK66N","Corey Ellis","01M74PNNJ1JR5BJQG3JK95RVFT",0,"MIT","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fcreate-mcp-server\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: create-mcp-server\ndescription: Build, iterate, and evaluate Model Context Protocol (MCP) servers that expose external services as tools an LLM can call. Use when asked to \"build an MCP server\", \"create an MCP tool\", \"wrap this API as MCP\", \"expose X to Claude\", or when extending OTTO with custom tool integrations. Covers research, schema\u002Ftool design, error handling, pagination, testing via MCP Inspector, and producing a 10-question eval set that proves the server actually enables real work.\n---\n\n\u003Cobjective>\nProduce a high-quality MCP server that an LLM can actually use — not one that merely parses spec-compliant. Quality is measured by how well the server enables real-world task completion, which means the tool descriptions, error messages, and pagination behave under model reasoning, not just at the wire level.\n\u003C\u002Fobjective>\n\n\u003Ccontext>\nthe agent consumes MCP heavily — see `src\u002Fresources\u002Fextensions\u002Fmcp-client\u002F`, `src\u002Fresources\u002Fextensions\u002Fotto\u002Fmcp-project-config.ts`, `src\u002Fresources\u002Fextensions\u002Fotto\u002Fworkflow-mcp.ts`, and `\u002Fotto mcp` commands. Users frequently want to extend OTTO with project-specific MCP servers (internal APIs, data sources, domain tools). This skill fills the authoring gap between \"MCP exists\" and \"I have a working server.\"\n\nInvocation points:\n- User describes a service or API they want an LLM to reach\n- `\u002Fotto mcp init` scaffolds config but there's a tool integration to build\n- Replacing a hand-rolled extension with a standard MCP server\n\u003C\u002Fcontext>\n\n\u003Ccore_principle>\n**THE QUALITY METRIC IS TASK COMPLETION, NOT SCHEMA VALIDITY.** A server that lists 30 tools with cryptic names and empty descriptions passes the protocol but fails the point. The tool description is the only thing an LLM has to decide whether to call it — write it like documentation for a stranger under time pressure.\n\n**DESIGN FOR THE MODEL, NOT THE API.** A raw REST endpoint is rarely the right tool. Group, filter, and pre-shape responses so the model gets what it needs to reason, not a 40KB JSON blob it has to summarize. Fewer, deeper tools beat many, shallow ones.\n\u003C\u002Fcore_principle>\n\n\u003Cprocess>\n\n## Step 1: Research and scope\n\n1. **Study modern MCP design.** Read the latest MCP protocol docs (not training data — fetch them). Read 2–3 reference implementations to see current patterns.\n2. **Pick a framework.** TypeScript is the default — the reference SDK is the most mature. Python is fine for data-heavy or ML adjacencies.\n3. **Analyze the target API.** Map the external service's endpoints, auth, rate limits, pagination, error shapes. Identify what a human workflow on top of it actually looks like — that's the cut line for tool design.\n4. **Produce a brief.** One page: what the server does, who calls it, the 5–10 tools you plan to expose, and the top 3 design trade-offs. Confirm with the user.\n\n## Step 2: Set up the project\n\nSkeleton:\n\n```text\nserver\u002F\n  src\u002F\n    index.ts         # MCP entry point — stdio or sse transport\n    client.ts        # API client with auth, retries, typed errors\n    tools\u002F           # one file per tool, or grouped by domain\n    pagination.ts    # shared cursor handling\n    errors.ts        # MCP-friendly error formatting\n  package.json       # @modelcontextprotocol\u002Fsdk as dep\n  tsconfig.json\n  README.md          # how to run, env vars, rate-limit notes\n  evals.xml          # 10 eval questions (Phase 4)\n```\n\nCore infrastructure goes first: API client with typed errors, pagination helpers, consistent retry\u002Ftimeout behavior. Do not inline these per tool.\n\n## Step 3: Implement tools\n\nFor each tool:\n\n1. **Name:** verb-noun, lowercase, snake_case. `search_issues`, `get_customer`, `create_deployment`. Not `do_thing` or `api_v2_post`.\n2. **Description (frontmatter):** 2–4 sentences. State what the tool does, when to use it, when NOT to use it, and any required fields or quirks. This is the model's entire interface to the tool — write it carefully.\n3. **Input schema (JSON Schema):** required fields marked, every field has a description, enums enumerated, examples included for free-form strings.\n4. **Output shape:** typed, minimal, decision-ready. If the raw API returns 40 fields and only 6 matter for follow-up calls, return 6.\n5. **Error handling:** never return raw HTTP errors. Translate to human-readable messages: \"Rate limit exceeded (retry in 30s)\", \"Authorization expired\", \"No record found for ID X\". Include the action the caller should take next.\n6. **Pagination:** expose cursors explicitly. Do not leak \"page N of M\" into the model — leak \"more results available, pass `cursor: abc123` to continue.\"\n\n## Step 4: Build and test with MCP Inspector\n\n1. Run the server under MCP Inspector. Verify it registers, every tool lists with its description, inputs schema-validate, outputs shape correctly.\n2. Call every tool at least once manually through the Inspector UI. Check error paths.\n3. Fix any \"looks fine in isolation, breaks under the Inspector's framing\" issues.\n\n## Step 5: Produce the eval set\n\nWrite 10 evaluation questions in `evals.xml` that exercise the server end-to-end. Each question should require 2+ tool calls and at least one decision the model has to make based on earlier output. Cover:\n\n- Happy path (2–3 questions)\n- Error recovery (2 questions — \"the first call failed, what next?\")\n- Pagination (1 question)\n- Decision under partial information (2–3 questions)\n- Cross-tool composition (1–2 questions)\n\nFormat:\n\n```xml\n\u003Cevals>\n  \u003Ceval id=\"1\">\n    \u003Cquestion>...user request...\u003C\u002Fquestion>\n    \u003Cexpected>...concrete observable answer or tool-call sequence...\u003C\u002Fexpected>\n  \u003C\u002Feval>\n\u003C\u002Fevals>\n```\n\nRun the evals. If the model can't complete them, the server — not the model — needs work. Iterate on descriptions, error messages, and tool granularity.\n\n## Step 6: Wire into OTTO\n\nWrite the project's `.mcp.json` entry using `\u002Fotto mcp init` as a starting point. Document env vars and startup in README.md. If the server is globally useful, suggest the user file it as a durable skill via `spike-wrap-up` or publish it.\n\n\u003C\u002Fprocess>\n\n\u003Canti_patterns>\n\n- **One-to-one REST mapping.** If the API has 30 endpoints, you likely want 6 tools.\n- **Empty or auto-generated descriptions.** \"Calls the \u002Fusers endpoint\" tells the model nothing it can reason with.\n- **Raw error passthrough.** `{\"error\": \"500\"}` is useless. Translate.\n- **Page-based pagination leaked as \"page 1 of 5\".** Use opaque cursors.\n- **Skipping the Inspector.** If you didn't run it under the Inspector, you didn't test it.\n- **No evals.** Without evals you have no signal on whether real task completion works.\n\n\u003C\u002Fanti_patterns>\n\n\u003Csuccess_criteria>\n\n- [ ] Every tool has a description that could guide a cold-start model correctly.\n- [ ] Errors are translated to actionable, human-readable messages.\n- [ ] Pagination uses opaque cursors; no leaked page numbers.\n- [ ] `evals.xml` has 10 questions; the model completes ≥8 without handholding.\n- [ ] MCP Inspector test passes cleanly.\n- [ ] README documents env, startup, and rate-limit behavior.\n- [ ] Server is reachable from OTTO via `.mcp.json` entry.\n\n\u003C\u002Fsuccess_criteria>\n","\u003Cobjective>\nProduce a high-quality MCP server that an LLM can actually use — not one that merely parses spec-compliant. Quality is measured by how well the server enables real-world task completion, which means the tool descriptions, error messages, and pagination behave under model reasoning, not just at the wire level.\n\u003C\u002Fobjective>\n\n\u003Ccontext>\nthe agent consumes MCP heavily — see `src\u002Fresources\u002Fextensions\u002Fmcp-client\u002F`, `src\u002Fresources\u002Fextensions\u002Fotto\u002Fmcp-project-config.ts`, `src\u002Fresources\u002Fextensions\u002Fotto\u002Fworkflow-mcp.ts`, and `\u002Fotto mcp` commands. Users frequently want to extend OTTO with project-specific MCP servers (internal APIs, data sources, domain tools). This skill fills the authoring gap between \"MCP exists\" and \"I have a working server.\"\n\nInvocation points:\n- User describes a service or API they want an LLM to reach\n- `\u002Fotto mcp init` scaffolds config but there's a tool integration to build\n- Replacing a hand-rolled extension with a standard MCP server\n\u003C\u002Fcontext>\n\n\u003Ccore_principle>\n**THE QUALITY METRIC IS TASK COMPLETION, NOT SCHEMA VALIDITY.** A server that lists 30 tools with cryptic names and empty descriptions passes the protocol but fails the point. The tool description is the only thing an LLM has to decide whether to call it — write it like documentation for a stranger under time pressure.\n\n**DESIGN FOR THE MODEL, NOT THE API.** A raw REST endpoint is rarely the right tool. Group, filter, and pre-shape responses so the model gets what it needs to reason, not a 40KB JSON blob it has to summarize. Fewer, deeper tools beat many, shallow ones.\n\u003C\u002Fcore_principle>\n\n\u003Cprocess>\n\n## Step 1: Research and scope\n\n1. **Study modern MCP design.** Read the latest MCP protocol docs (not training data — fetch them). Read 2–3 reference implementations to see current patterns.\n2. **Pick a framework.** TypeScript is the default — the reference SDK is the most mature. Python is fine for data-heavy or ML adjacencies.\n3. **Analyze the target API.** Map the external service's endpoints, auth, rate limits, pagination, error shapes. Identify what a human workflow on top of it actually looks like — that's the cut line for tool design.\n4. **Produce a brief.** One page: what the server does, who calls it, the 5–10 tools you plan to expose, and the top 3 design trade-offs. Confirm with the user.\n\n## Step 2: Set up the project\n\nSkeleton:\n\n```text\nserver\u002F\n  src\u002F\n    index.ts         # MCP entry point — stdio or sse transport\n    client.ts        # API client with auth, retries, typed errors\n    tools\u002F           # one file per tool, or grouped by domain\n    pagination.ts    # shared cursor handling\n    errors.ts        # MCP-friendly error formatting\n  package.json       # @modelcontextprotocol\u002Fsdk as dep\n  tsconfig.json\n  README.md          # how to run, env vars, rate-limit notes\n  evals.xml          # 10 eval questions (Phase 4)\n```\n\nCore infrastructure goes first: API client with typed errors, pagination helpers, consistent retry\u002Ftimeout behavior. Do not inline these per tool.\n\n## Step 3: Implement tools\n\nFor each tool:\n\n1. **Name:** verb-noun, lowercase, snake_case. `search_issues`, `get_customer`, `create_deployment`. Not `do_thing` or `api_v2_post`.\n2. **Description (frontmatter):** 2–4 sentences. State what the tool does, when to use it, when NOT to use it, and any required fields or quirks. This is the model's entire interface to the tool — write it carefully.\n3. **Input schema (JSON Schema):** required fields marked, every field has a description, enums enumerated, examples included for free-form strings.\n4. **Output shape:** typed, minimal, decision-ready. If the raw API returns 40 fields and only 6 matter for follow-up calls, return 6.\n5. **Error handling:** never return raw HTTP errors. Translate to human-readable messages: \"Rate limit exceeded (retry in 30s)\", \"Authorization expired\", \"No record found for ID X\". Include the action the caller should take next.\n6. **Pagination:** expose cursors explicitly. Do not leak \"page N of M\" into the model — leak \"more results available, pass `cursor: abc123` to continue.\"\n\n## Step 4: Build and test with MCP Inspector\n\n1. Run the server under MCP Inspector. Verify it registers, every tool lists with its description, inputs schema-validate, outputs shape correctly.\n2. Call every tool at least once manually through the Inspector UI. Check error paths.\n3. Fix any \"looks fine in isolation, breaks under the Inspector's framing\" issues.\n\n## Step 5: Produce the eval set\n\nWrite 10 evaluation questions in `evals.xml` that exercise the server end-to-end. Each question should require 2+ tool calls and at least one decision the model has to make based on earlier output. Cover:\n\n- Happy path (2–3 questions)\n- Error recovery (2 questions — \"the first call failed, what next?\")\n- Pagination (1 question)\n- Decision under partial information (2–3 questions)\n- Cross-tool composition (1–2 questions)\n\nFormat:\n\n```xml\n\u003Cevals>\n  \u003Ceval id=\"1\">\n    \u003Cquestion>...user request...\u003C\u002Fquestion>\n    \u003Cexpected>...concrete observable answer or tool-call sequence...\u003C\u002Fexpected>\n  \u003C\u002Feval>\n\u003C\u002Fevals>\n```\n\nRun the evals. If the model can't complete them, the server — not the model — needs work. Iterate on descriptions, error messages, and tool granularity.\n\n## Step 6: Wire into OTTO\n\nWrite the project's `.mcp.json` entry using `\u002Fotto mcp init` as a starting point. Document env vars and startup in README.md. If the server is globally useful, suggest the user file it as a durable skill via `spike-wrap-up` or publish it.\n\n\u003C\u002Fprocess>\n\n\u003Canti_patterns>\n\n- **One-to-one REST mapping.** If the API has 30 endpoints, you likely want 6 tools.\n- **Empty or auto-generated descriptions.** \"Calls the \u002Fusers endpoint\" tells the model nothing it can reason with.\n- **Raw error passthrough.** `{\"error\": \"500\"}` is useless. Translate.\n- **Page-based pagination leaked as \"page 1 of 5\".** Use opaque cursors.\n- **Skipping the Inspector.** If you didn't run it under the Inspector, you didn't test it.\n- **No evals.** Without evals you have no signal on whether real task completion works.\n\n\u003C\u002Fanti_patterns>\n\n\u003Csuccess_criteria>\n\n- [ ] Every tool has a description that could guide a cold-start model correctly.\n- [ ] Errors are translated to actionable, human-readable messages.\n- [ ] Pagination uses opaque cursors; no leaked page numbers.\n- [ ] `evals.xml` has 10 questions; the model completes ≥8 without handholding.\n- [ ] MCP Inspector test passes cleanly.\n- [ ] README documents env, startup, and rate-limit behavior.\n- [ ] Server is reachable from OTTO via `.mcp.json` entry.\n\n\u003C\u002Fsuccess_criteria>\n",{"name":5,"description":7},"main","e62d23e713a36a0570f7f5c3ee69868b1131e548","src\u002Fresources\u002Fskills\u002Fcreate-mcp-server","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli","https:\u002F\u002Fgithub.com\u002Fcmetech","https:\u002F\u002Fwww.cmetech.io","Corey Ellis is a seasoned data engineer with over 20 years of experience in the field. He is the founder of CME Technologies LLC.","2026-06-17 22:50:56",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add cmetech\u002Fotto-cli --skill create-mcp-server",{"tool":38,"command":39},"bunx","bunx skills add cmetech\u002Fotto-cli --skill create-mcp-server",{"tool":41,"command":42},"pnpm","pnpm dlx skills add cmetech\u002Fotto-cli --skill create-mcp-server","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fcreate-mcp-server\u002FSKILL.md",true,[],[47,54,61,68,75,82],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M74PWE6AVSD8RDS1M3Q6TH95","write-milestone-brief","Synthesize the current conversation into a milestone brief (PRD). Writes to `M###-CONTEXT.md` by default, or files a GitHub issue only with explicit user confirmation. Use when asked to \"turn this into a PRD\", \"draft a milestone brief\", \"capture this context\", \"write it up\", or when enough has been discussed to commit the plan to paper. Does not interview — it synthesizes what is already known.","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fwrite-milestone-brief\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M74PWB6CEEQPGS9HRQY4QQPZ","write-docs","Collaborative document authoring workflow for proposals, technical specs, decision docs, README sections, ADRs, and long-form prose that must work for fresh readers. Use when asked to \"write the docs\", \"draft a proposal\", \"write a spec\", \"write an RFC\", \"write the README\", or when a document needs to be understandable by someone without this session's context. Three stages: gather context, iterate on structure, reader-test for a stranger.","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fwrite-docs\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M74PW7SJQ8H516MDBQ7WV1BN","web-quality-audit","Comprehensive web quality audit covering performance, accessibility, SEO, and best practices. Use when asked to \"audit my site\", \"review web quality\", \"run lighthouse audit\", \"check page quality\", or \"optimize my website\".","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fweb-quality-audit\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M74PW5JAA6D34Z8QBN1HQ014","web-design-guidelines","Review UI code for Web Interface Guidelines compliance. Use when asked to \"review my UI\", \"check accessibility\", \"audit design\", \"review UX\", or \"check my site against best practices\".","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fweb-design-guidelines\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M74PW1YP2Y03GMDYD443BVHQ","verify-before-complete","Block completion claims until verification evidence has been produced in the current message. Use before marking a task\u002Fslice\u002Fmilestone complete, before creating a commit or PR, before saying \"it works\" or \"tests pass\", and any time you are about to claim work is done. The rule is: evidence before claims, always — running the verification must happen now, not \"earlier in the session\". Fresh output or no claim.","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fverify-before-complete\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M74PVXR5GE8HE55YWEG7ENBP","userinterface-wiki","UI\u002FUX best practices for web interfaces. Use when reviewing animations, CSS, audio, typography, UX patterns, prefetching, or icon implementations. Covers 11 categories from animation principles to typography. Outputs file:line findings.","https:\u002F\u002Fgithub.com\u002Fcmetech\u002Fotto-cli\u002Fblob\u002Fmain\u002Fsrc\u002Fresources\u002Fskills\u002Fuserinterface-wiki\u002FSKILL.md",[],[]]