For agents
The Dialect Engineering connector
An MCP server that helps an agent explain how the paper SaaS architecture when code is cheap applies to a person's own product.
What it does
The connector gives an agent the paper's knowledge graph: the layers of a SaaS product, the symptoms a provider sees today, the principles that address them, the recommendations for new and existing products, and the limits the paper states, each tied to the passages of the paper and the moments in its explainer video that support it.
With it, an agent can interview a person about their product, map what they describe onto the paper, and write an explanation in a small language built on the graph. The connector checks that explanation against the graph and the paper's rules, and returns a link. Opening the link plays the explanation on this site as a short video: the recorded narration explains the paper, and the agent's lines connect it to the person's question.
Add it
Server address: https://dialect-engineering.ai/mcp. It uses Streamable HTTP, needs no sign-in or account, and every tool is read-only.
- Claude (web, desktop and mobile): in Settings, under Connectors, add a custom connector with the address above.
- Claude Code:
claude mcp add --transport http dialect-engineering https://dialect-engineering.ai/mcp - Other MCP clients: add a remote server over Streamable HTTP with the address above.
- Agents in the browser: where a browser supports WebMCP, the site offers the same tools directly, with nothing to add.
Tools
| Tool | What it does |
|---|---|
explanation_guide | Returns the reference for the explanation language: how it is meant to be used, its statements, worked examples, and the full knowledge graph of the paper 'SaaS architecture when code is cheap'. |
graph_concepts | Concepts from the paper's knowledge graph, optionally filtered by kind (context, layer, demand, cause, symptom, principle, recommendation, limit, case) and, for symptoms, by layer. |
graph_links | Every link the paper makes to or from one concept: what addresses a symptom, what a principle requires, what limits it, which context a recommendation applies to. |
concept_evidence | Returns the video sentences and paper paragraphs behind one concept, or behind the link between two concepts. |
find_in_paper | Keyword search over the video's narration and the paper's paragraphs. Returns the best-matching sentences and paragraphs with their addresses. |
check_explanation | Checks an explanation written in the explanation language against the knowledge graph and the paper's rules. Returns each problem with its line and a reason, and what it compiles to. |
make_explanation | Checks an explanation and, if it passes, returns a link that opens it on this site in the person's own browser, where it plays and is saved there. The explanation travels in the link itself; the server keeps nothing. |
The language and the whole graph are also published at /explanation-language.md, and the graph with its evidence at /graph.
Example prompts
- Help me understand how the dialect engineering paper applies to our onboarding. It takes us months and customers find configuration errors after go-live.
- We sell a payroll product priced by tier, and customers say they pay for features they never use. What does the paper suggest?
- We are starting a new HR product from scratch. What would the paper have us design differently?
- Our customers keep asking for changes that end up as special cases in shared code. How does the paper see that problem, and where would we start?
Privacy and support
The connector keeps no data and needs no account. See the privacy policy. For questions or problems, contact hello@dialect-engineering.ai.