# OwnerSpec for AI agents: MCP server, A2A agent and discovery files > Connect an AI assistant to OwnerSpec's cited answers: an MCP server at /mcp with 8 read-only tools and no authentication, plus A2A, Markdown and discovery files. | | | |---|---| | Source | https://ownerspec.com/mcp-server/ | | Publisher | OwnerSpec (https://ownerspec.com/) | | Published | 2026-09-12 | | Updated | 2026-09-12 | --- OwnerSpec publishes its content for agents the same way it publishes it for people: the figure, the formula and the source, with the canonical URL and the date the facts were last checked. This page is the technical reference for connecting an assistant to it. ## Two layers: what to cite and what to call **The public citation layer** is the set of canonical OwnerSpec web pages, one per fact (`https://ownerspec.com////`). Each page carries a short answer, a fact sheet (answer, formula or compatibility basis, inputs, assumptions, method, primary source), a facts-verified date, a typed source list and its own canonical URL. That URL is the citation target. **The machine-readable layer** is how a compatible agent retrieves the same content in a cheaper shape: the Markdown twin of every page, `/llms.txt`, `/llms-full.txt`, `/index.json`, the MCP server at `/mcp` and the A2A agent at `/a2a`. These are retrieval interfaces. When information is retrieved through any of them, the canonical public page should be used as the citation target whenever one exists; never cite the endpoint or the Markdown URL. Every tool response and every Markdown twin therefore carries, for each page: the canonical URL, the page title, the facts-verified date, the source list, the fact sheet, and the tool or representation it came from. Nothing on this page implies that publishing these interfaces causes any assistant, search engine or answer engine to index or cite OwnerSpec; they make retrieval and correct citation possible, and no more. ## Endpoint The MCP (Model Context Protocol) server is `https://ownerspec.com/mcp`. It is a stateless Streamable HTTP server: JSON-RPC 2.0 over POST, one JSON response per request, no sessions, no SSE stream, no authentication. GET returns 405 by design. ## Adding it to a client | Client | Steps | | --- | --- | | Claude (web or desktop) | Settings, Connectors, Add custom connector, paste `https://ownerspec.com/mcp`, choose "No authentication" | | Claude Code | `claude mcp add --transport http ownerspec https://ownerspec.com/mcp` | | ChatGPT | Settings, Connectors, Advanced, Developer mode, add `https://ownerspec.com/mcp`, no authentication | | Perplexity (Pro, Max, Enterprise) | Settings, Connectors, Custom connector, Remote, paste the endpoint, authentication "None" | | MCP Inspector | `npx @modelcontextprotocol/inspector https://ownerspec.com/mcp` | ## The tools | Tool | What it returns | | --- | --- | | `search_pages` | Full-text search over every page, filterable by layer (guides, calculators, parts, reviews) and topic | | `get_page` | One page as clean Markdown with its canonical URL, dates and sources | | `get_quick_answer` | The page's cited quick answer (134 to 167 words), the facts-verified date and the source list: the passage to quote | | `diagnose_water_problem` | A symptom or lab result in, the matching diagnosis guides and their quick answers out | | `find_replacement_part` | A model or part number in, the replacement cartridges, lamps or parts that fit, with product links | | `get_product_picks` | The picks from the review and parts pages that match a need, with Amazon product links | | `convert_water_hardness` | Any hardness reading converted to gpg, mg/L, dH, fH, Clark and mmol/L with the USGS band | | `size_water_softener` | Grain capacity between regenerations from people, use, hardness, iron and days, with the nominal size and the salt-efficient Fleck 5600SXT row | Every tool answer ends with the URL to cite and, where the page carries one, the date its facts were verified. `get_quick_answer` also returns the page's fact sheet verbatim: answer, formula or compatibility basis, inputs, output, assumptions, method and the typed primary source. The two calculator tools use exactly the formulas and constants printed on the calculator pages, so a number from the tool and a number from the page always agree (`scripts/verify-build.mjs` fails the build if they drift). The server also exposes resources (`llms.txt`, `llms-full.txt`, the sitemap, the search index, the OpenAPI description, `auth.md`) and one prompt, `answer_from_ownerspec`, that walks an assistant through search, quick answer and citation. ## Example prompts - "My well water smells like rotten eggs. Which treatment class handles that?" (diagnose, then cite the guide) - "Which filters fit an iSpring RCC7AK?" (part match with product links) - "Size a softener for 4 people at 18 gpg with 1 mg/L iron, regenerating weekly." - "Convert 14 dH to grains per gallon and tell me if that is hard water." ## What it does not do It does not sell, install or service anything, and it never fetches live prices or stock. Product links go to the retailer's ordinary product page. It gives no medical advice: contaminant figures are EPA, WHO or state limits quoted from the page that cites them. ## Limits Read-only, public, no rate limit published. Be a good citizen: identify your client with a descriptive User-Agent and cache `index.json` rather than searching a hundred times a minute. ## A2A agent An A2A (Agent-to-Agent) endpoint at `https://ownerspec.com/a2a` accepts a plain-language question in a `SendMessage` (or `message/send`) call and returns a completed task whose artifact holds the best-matching pages, their quick answers and the URLs to cite. The agent card is `/.well-known/agent-card.json`. ## Affiliate disclosure OwnerSpec is funded by affiliate commission. Product links returned by `find_replacement_part` and `get_product_picks` are Amazon Associates links: a purchase may earn OwnerSpec a commission at no extra cost to the buyer. Every response that carries such a link says so, and an assistant presenting them should say so once too. Editorial content is free to read, with no paywall and no account. ## Discovery files | File | Purpose | | --- | --- | | `/.well-known/mcp/server-card.json` | MCP server card (SEP-1649) | | `/.well-known/agent-card.json` | A2A agent card | | `/.well-known/agent-skills/index.json` | Agent Skills discovery index with the `ownerspec` skill | | `/.well-known/api-catalog` | RFC 9727 API catalog | | `/.well-known/ard.json` | Agentic Resource Discovery manifest | | `/openapi.json` | OpenAPI 3.1 description of the read-only endpoints | | `/auth.md` | Authentication policy: none required | | `/llms.txt`, `/llms-full.txt` | Site index and full text for language models | | `/index.json` | The search index the tools read | Every page is also served as Markdown: append `index.md` to its URL, or send `Accept: text/markdown`. ## Source code The server is a single Cloudflare Pages Function, published under the MIT licence at [github.com/ownerspec-com/mcp-server](https://github.com/ownerspec-com/mcp-server). ## Policies [Editorial policy](/legal/editorial-policy/), [affiliate disclosure](/legal/affiliate-disclosure/), [terms of use](/legal/terms/) (quoting with attribution and a link is permitted), [privacy](/legal/privacy/). Questions: support@voyagehacks.com. --- Citation: OwnerSpec, "OwnerSpec for AI agents: MCP server, A2A agent and discovery files", https://ownerspec.com/mcp-server/, updated 12 September 2026. Canonical citation target: the URL above (the public HTML page), not this Markdown representation. Reuse: quoting with attribution and a link is permitted. See https://ownerspec.com/legal/terms/.