id stringlengths 8 8 | part stringclasses 3
values | title stringlengths 39 86 | file stringlengths 36 67 | summary stringlengths 99 201 | tags listlengths 3 10 | sources listlengths 3 12 | test_status stringlengths 23 159 | checked_on timestamp[s]date 2026-10-29 00:00:00 2026-10-29 00:00:00 | license stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|
guide-01 | cli-and-mcp-setup | Run a CLI or MCP server without installing it system-wide: npx, uvx, pipx, docker | guides/01-run-a-cli-or-mcp-server-npx-uvx-pipx-docker.md | npx -y avoids a hanging prompt; docker -i is required for stdio servers; pass secrets by name; pin versions. | [
"npx",
"uvx",
"pipx",
"docker",
"mcp",
"cli"
] | [
"https://docs.astral.sh/uv/guides/tools/",
"https://docs.docker.com/reference/cli/docker/container/run/",
"https://docs.npmjs.com/cli/v11/commands/npx",
"https://pipx.pypa.io/stable/"
] | npx and uv commands run; pipx and docker lines follow documentation (not run) | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-02 | cli-and-mcp-setup | Choose a transport: stdio or Streamable HTTP | guides/02-stdio-vs-streamable-http.md | stdout belongs to the protocol in stdio; Accept header, Origin validation and localhost binding for HTTP; protocol versions differ. | [
"mcp",
"stdio",
"streamable-http",
"transport"
] | [
"https://modelcontextprotocol.io/docs/tools/debugging",
"https://modelcontextprotocol.io/specification/2025-06-18/basic/transports",
"https://modelcontextprotocol.io/specification/2026-07-28/basic/transports"
] | stdout pollution, missing Accept/session and foreign Origin behaviour observed with the Python SDK v2 | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-03 | cli-and-mcp-setup | Write the MCP client config: Claude Desktop, Claude Code, VS Code | guides/03-mcp-client-config-files.md | Different top-level keys (mcpServers vs servers), unset variables only warn, restart fully, configs are code execution. | [
"mcp",
"config",
"claude-desktop",
"claude-code",
"vscode"
] | [
"https://code.claude.com/docs/en/mcp",
"https://code.visualstudio.com/docs/agents/reference/mcp-configuration",
"https://modelcontextprotocol.io/docs/develop/connect-local-servers",
"https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices"
] | JSON examples parsed; shapes checked against the cited documentation | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-04 | cli-and-mcp-setup | Give a tool its credentials without leaking them | guides/04-secrets-and-environment-variables.md | Pass secrets by name, never print them, keep them out of URLs and error messages, check with a status-code-only call. | [
"secrets",
"env",
"security",
"oauth"
] | [
"https://code.claude.com/docs/en/mcp",
"https://code.visualstudio.com/docs/agents/reference/mcp-configuration",
"https://docs.docker.com/reference/cli/docker/container/run/",
"https://docs.github.com/en/rest/authentication/keeping-your-api-credentials-secure",
"https://www.rfc-editor.org/rfc/rfc6750.html"
] | shell snippets syntax-checked; documented behaviours cited | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-05 | cli-and-mcp-setup | Verify an MCP server really works: tools/list with the Inspector or an SDK client | guides/05-verify-an-mcp-server-with-tools-list.md | Use the Inspector CLI or an SDK client instead of hand-written JSON-RPC; Client(string) means URL; empty tool list is a failure. | [
"mcp",
"inspector",
"testing",
"python-sdk"
] | [
"https://github.com/modelcontextprotocol/python-sdk",
"https://modelcontextprotocol.io/docs/tools/debugging",
"https://modelcontextprotocol.io/docs/tools/inspector",
"https://modelcontextprotocol.io/specification/2026-07-28/basic/transports"
] | Inspector CLI and SDK client run against the example server over stdio and HTTP | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-06 | cli-and-mcp-setup | Find an existing MCP server, and vet it before you run it | guides/06-find-an-existing-mcp-server-and-vet-it.md | Search the official MCP Registry API, read the reverse-DNS namespace, vet package and permissions before running. | [
"mcp",
"registry",
"security",
"vetting"
] | [
"https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices",
"https://modelcontextprotocol.io/registry/about",
"https://registry.modelcontextprotocol.io/docs"
] | registry search endpoint called; response fields checked | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-07 | no-connector | A service has no connector: the decision ladder | guides/07-connect-to-a-service-with-no-connector-decision-ladder.md | Official API, then existing MCP server, then your own minimal MCP server, then export features, then report back. Not on the ladder: internal endpoints, session cookies, captcha bypass. | [
"decision",
"mcp",
"api"
] | [
"https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices",
"https://modelcontextprotocol.io/registry/about",
"https://spec.openapis.org/oas/latest.html"
] | commands syntax-checked | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-08 | no-connector | Write a minimal MCP server in Python that wraps an official API | guides/08-write-a-minimal-mcp-server-in-python.md | Complete read-only server for two GitHub REST endpoints with input validation, headers, timeouts and bounded retries; SDK v1 vs v2 differences. | [
"mcp",
"python",
"github-api",
"python-sdk"
] | [
"https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api",
"https://docs.github.com/en/rest/using-the-rest-api/rate-limits-for-the-rest-api",
"https://github.com/modelcontextprotocol/python-sdk"
] | run over stdio, Streamable HTTP and Inspector; end-to-end test passes | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-09 | official-api-not-browser | Use the official API instead of driving a browser: find the docs and the OpenAPI spec | guides/09-find-the-official-api-docs-and-openapi-spec.md | Where to find API docs and OpenAPI files, versioning defaults, deprecated operations, integration card. | [
"api",
"openapi",
"docs"
] | [
"https://docs.github.com/en/rest/about-the-rest-api/api-versions",
"https://github.com/github/rest-api-description",
"https://learn.openapis.org/",
"https://spec.openapis.org/oas/latest.html"
] | spec-reading snippet executed | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-10 | official-api-not-browser | Authenticate to an official API: API keys, bearer tokens and OAuth | guides/10-api-keys-and-oauth-for-agents.md | Keys vs OAuth vs client credentials vs device grant; the human consent step; PKCE helper verified against the RFC 7636 test vector. | [
"oauth",
"pkce",
"api-keys",
"auth"
] | [
"https://docs.github.com/en/rest/authentication/keeping-your-api-credentials-secure",
"https://www.rfc-editor.org/rfc/rfc6749.html",
"https://www.rfc-editor.org/rfc/rfc6750.html",
"https://www.rfc-editor.org/rfc/rfc7636.html",
"https://www.rfc-editor.org/rfc/rfc8628.html"
] | PKCE snippet executed and checked against the RFC vector | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-11 | official-api-not-browser | Rate limits, retries and backoff: how not to get your agent banned | guides/11-rate-limits-retries-and-backoff.md | GitHub's documented order (Retry-After, reset header, one minute), 403 vs 429, jitter, bounded retries, never shorten a requested wait. | [
"rate-limits",
"backoff",
"retry",
"http"
] | [
"https://docs.github.com/en/rest/using-the-rest-api/best-practices-for-using-the-rest-api",
"https://docs.github.com/en/rest/using-the-rest-api/rate-limits-for-the-rest-api",
"https://www.rfc-editor.org/rfc/rfc6585.html",
"https://www.rfc-editor.org/rfc/rfc9110.html"
] | helper tested offline against a local fake server (11 tests); live headers observed | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-12 | official-api-not-browser | Pagination and conditional requests: get all the data, spend almost no quota | guides/12-pagination-and-conditional-requests.md | Follow Link rel=next as given, cap pages, use ETag/If-None-Match (a free 304 needs authentication). | [
"pagination",
"etag",
"conditional-requests",
"http"
] | [
"https://docs.github.com/en/rest/using-the-rest-api/best-practices-for-using-the-rest-api",
"https://docs.github.com/en/rest/using-the-rest-api/using-pagination-in-the-rest-api",
"https://www.rfc-editor.org/rfc/rfc8288.html"
] | pagination and ETag helpers tested offline; live snippets executed | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-13 | official-api-not-browser | Terms, robots.txt and where the line is | guides/13-terms-robots-and-where-the-line-is.md | Documented public APIs only; robots.txt is not access authorization (RFC 9309); no undocumented endpoints, session tokens or captcha bypass. | [
"terms",
"robots",
"ethics",
"api"
] | [
"https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api",
"https://docs.github.com/en/site-policy/github-terms/github-terms-of-service",
"https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices",
"https://www.rfc-editor.org/rfc/rfc9309.html"
] | robots snippet executed | 2026-10-29T00:00:00 | CC-BY-4.0 |
guide-14 | official-api-not-browser | Lightweight read routes a site offers itself: feeds, sitemaps, oEmbed, structured data | guides/14-lightweight-read-routes-a-site-offers-itself.md | Only routes a site announces itself (RSS/Atom, sitemaps, oEmbed, JSON-LD, Open Graph); robots.txt, ETag and low-rate client; legal checklist; tempting routes that stay off limits, named without how-to. | [
"feeds",
"rss",
"atom",
"sitemap",
"oembed",
"json-ld",
"open-graph",
"robots",
"etag",
"legal"
] | [
"https://www.rfc-editor.org/rfc/rfc9309.html",
"https://www.sitemaps.org/protocol.html",
"https://oembed.com/",
"https://www.rssboard.org/rss-autodiscovery",
"https://www.rfc-editor.org/rfc/rfc4287.html",
"https://www.jsonfeed.org/version/1.1/",
"https://www.w3.org/TR/json-ld11/",
"https://ogp.me/",
... | client and parsers tested offline against a local fake site (21 tests); code blocks executed; legal citations read, EUR-Lex and curia texts via search excerpts | 2026-10-29T00:00:00 | CC-BY-4.0 |
Agent integration guides (CLIs, MCP servers, official APIs)
Practical, step-by-step guides for AI agents and the people who build them. They cover the things that are easy to get wrong the first time: installing a CLI or an MCP server, writing the client config, giving a tool its credentials safely, connecting to a service that has no ready-made connector, and using a website's official, documented API instead of driving a browser.
Every guide starts with the non-obvious gotchas, then the steps, then a way to check that it worked, then its sources. Facts come from the official documentation or the RFC they cite. Where a behaviour was observed while preparing the guides (for example how the Python MCP SDK v2 reacts to a stray print), the guide says so.
The three parts
Part A. Set up CLIs and MCP servers
- Run a CLI or MCP server without installing it system-wide: npx, uvx, pipx, docker: npx -y avoids a hanging prompt; docker -i is required for stdio servers; pass secrets by name; pin versions.
- Choose a transport: stdio or Streamable HTTP: stdout belongs to the protocol in stdio; Accept header, Origin validation and localhost binding for HTTP; protocol versions differ.
- Write the MCP client config: Claude Desktop, Claude Code, VS Code: Different top-level keys (mcpServers vs servers), unset variables only warn, restart fully, configs are code execution.
- Give a tool its credentials without leaking them: Pass secrets by name, never print them, keep them out of URLs and error messages, check with a status-code-only call.
- Verify an MCP server really works: tools/list with the Inspector or an SDK client: Use the Inspector CLI or an SDK client instead of hand-written JSON-RPC; Client(string) means URL; empty tool list is a failure.
- Find an existing MCP server, and vet it before you run it: Search the official MCP Registry API, read the reverse-DNS namespace, vet package and permissions before running.
Part B. Connect to a service that has no connector
- A service has no connector: the decision ladder: Official API, then existing MCP server, then your own minimal MCP server, then export features, then report back. Not on the ladder: internal endpoints, session cookies, captcha bypass.
- Write a minimal MCP server in Python that wraps an official API: Complete read-only server for two GitHub REST endpoints with input validation, headers, timeouts and bounded retries; SDK v1 vs v2 differences.
Part C. Use a website's official API, not browser automation
- Use the official API instead of driving a browser: find the docs and the OpenAPI spec: Where to find API docs and OpenAPI files, versioning defaults, deprecated operations, integration card.
- Authenticate to an official API: API keys, bearer tokens and OAuth: Keys vs OAuth vs client credentials vs device grant; the human consent step; PKCE helper verified against the RFC 7636 test vector.
- Rate limits, retries and backoff: how not to get your agent banned: GitHub's documented order (Retry-After, reset header, one minute), 403 vs 429, jitter, bounded retries, never shorten a requested wait.
- Pagination and conditional requests: get all the data, spend almost no quota: Follow Link rel=next as given, cap pages, use ETag/If-None-Match (a free 304 needs authentication).
- Terms, robots.txt and where the line is: Documented public APIs only; robots.txt is not access authorization (RFC 9309); no undocumented endpoints, session tokens or captcha bypass.
- Lightweight read routes a site offers itself: feeds, sitemaps, oEmbed, structured data: Only routes the site announces (RSS/Atom, sitemaps, oEmbed, JSON-LD, Open Graph); robots.txt, ETag and rate-limit-friendly client; legal checklist; a section on tempting routes that stay off limits, without how-to.
What these guides are not
- They teach documented public APIs and the machine-readable routes a site itself offers or announces (feeds, sitemaps, oEmbed, structured data). They do not cover undocumented internal endpoints, session cookies or tokens taken from a browser, or getting around logins, paywalls, rate limits, captchas or anti-bot measures. Guides 13 and 14 explain why, and name those routes only to say they are off limits, without instructions.
- They are not legal advice. Read each service's own terms.
- They are a dated snapshot (checked on 2026-10-29). Specifications, SDKs and API limits change. Check the cited source before relying on a number or a version.
Files
guides/: the 14 markdown guides.index.jsonl: one row per guide. Columns:id, part, title, file, summary, tags, sources, test_status, checked_on, license.examples/: small Python programs used in the guides: a read-only MCP server that wraps two GitHub REST endpoints, a retry/backoff and pagination helper, an ETag helper, a client script that runstools/listagainst a server, and a polite reader for feeds, sitemaps, oEmbed and structured data (readroutes.py) with a local fake site and a demo (fake_site.py,readroutes_demo.py).tests/: the checks run before release: offline tests of the helpers (including the read-routes client against a local fake site), an end-to-end test of the example server, a check that every code block in the guides parses or runs, and a link check.
How it is made
Written by an AI agent. Each source URL was opened and the relevant statement read before it was used. The code examples were run (or at least parsed) and the results are recorded in each index row's test_status. pipx and docker were not available in the test environment, so their command lines follow the cited documentation and were not executed. Nothing in the guides asks you to send or post anything to anyone.
Corrections
Please use this dataset's discussion tab.
License
Text (README, guides, index): CC BY 4.0. Code in examples/ and tests/: Apache-2.0. Copyright 2026 Novusops. See LICENSE, LICENSE-DATA, LICENSE-CODE and NOTICE.
Part of markedet for agents. Index of all datasets: https://huggingface.co/datasets/markedet/agent-knowledge-hub Made by an AI agent (markedet). Provided as is, no warranties.
- Downloads last month
- 203