{
  "uid": "cogitave.learn.native-mcp-interface.introduction",
  "kind": "moduleUnit",
  "href": "/modules/native-mcp-interface/introduction/",
  "title": "Introduction",
  "summary": "",
  "type": null,
  "products": [],
  "roles": [],
  "levels": [],
  "subjects": [],
  "headings": [
    "What you will get from this module",
    "Where this fits"
  ],
  "source": "An AI-native organization keeps its whole estate in one canonical model - a\nproperty graph that every product, standard, and agent shares - and then has to\ndecide how humans and agents actually *reach* it. The transferable answer this\nmodule teaches: expose that model **natively over the Model Context Protocol\n(MCP)**, so a person's UI and an agent call the same tools, against the same\nschema, and get the same answer. Cogitave's Core is the worked example; earlier\nmodules in this path build and query that graph, and this module is about the\nsurface an agent actually calls to reach it: the **native MCP interface**.\n\n\"Native\" is a specific claim, not a decoration - and a design decision worth\nmaking deliberately. In Cogitave's estate,\n[ADR-0003](../../../../core/docs/decisions/0003-mcp-native.md) records the\ndecision in plain terms: Core's tools and resources over the Model Context\nProtocol are the **canonical** query surface, and REST, GraphQL, and `llms.txt`\nare generated conveniences over the identical catalog, graph, and search\nmaterializations - not the other way around. A human's UI and an agent call the\nsame tool, against the same schema, and get the same answer.\n\nConcretely, in Cogitave's instance, Core embeds an MCP server directly in its\nRust hot path, targeting **spec revision 2025-11-25**, reachable over stdio for\nlocal agents (the namzu kernel, CI) or Streamable HTTP for edge and remote\ncallers. Every tool's input and output validate against **JSON Schema 2020-12** -\nthe same dialect the property graph's own schema uses, so a tool's structured\nresult and the graph model agree by construction rather than by convention.\n\nEverything in this module is grounded in what\n[mcp-interface.md](../../../../core/docs/mcp-interface.md) and ADR-0003 actually\nstate today for that instance. Read this module the way you would read those\ndocuments: as the source of truth, with this module teaching you to use them\nrather than paraphrasing them from memory.\n\n## What you will get from this module\n\nBy the end, you will be able to reason about a call to a canonical model over\nMCP end to end - Cogitave's Core here: why the surface is native rather than\nadapted, what protocol guarantees you can rely on - transports, schema dialect,\nthe error model - which tools and resources exist and what each one returns, and\nwhere the governance tools stop at \"propose\" instead of \"mutate.\"\n\n## Where this fits\n\nThis is the third module of the path that teaches you to build on a single\ncanonical model. It assumes you can already picture such a model as a property\ngraph and does not re-teach that model; if a term here is unfamiliar,\n[mcp-interface.md](../../../../core/docs/mcp-interface.md) links onward to the\nsubstrate and query docs it builds on.\n",
  "partOf": "cogitave.learn.native-mcp-interface",
  "durationInMinutes": 4,
  "quiz": null
}