MCP Endpoint

The IKARY Manifest API is available as an MCP server. AI agents and MCP-compatible clients can discover schemas, validate manifests, and get structural guidance through tool calls over the MCP protocol.

Endpoint

POST https://public.ikary.co/mcp

The transport is Streamable HTTP (JSON-RPC 2.0). The endpoint is stateless. No session management or initialization handshake is required.

Connecting

Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "ikary-manifest": {
      "url": "https://public.ikary.co/mcp"
    }
  }
}

Cursor

Add this to your .cursor/mcp.json:

{
  "mcpServers": {
    "ikary-manifest": {
      "url": "https://public.ikary.co/mcp"
    }
  }
}

Claude Code

claude mcp add ikary-manifest https://public.ikary.co/mcp

Raw JSON-RPC

curl -X POST https://public.ikary.co/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/list" }'

Available tools

The server exposes 19 tools across five groups.

Discovery

Tool	Description	Parameters
get_manifest_schema	Returns the CellManifestV1 structure with fields, types, and semantic rules	version (string, optional)
get_entity_definition_schema	Returns the EntityDefinition contract	version (string, optional)
get_page_schema	Returns page types and their contracts	pageType (string, optional)
get_capability_schema	Returns the CapabilityDefinition contract	capabilityType (string, optional)

Registry

Tool	Description	Parameters
list_primitives	Returns the primitive catalog; pass source: "custom" for project-specific ones	category (string, optional), source (string, optional)
get_primitive_contract	Returns one primitive's full props contract	primitive (string, required)
list_examples	Returns available sample manifests	(none)
get_example_manifest	Returns the full content of one example	example (string, required)

Primitives

These tools have full capability when connecting to the local MCP server (ikary setup ai --local). With the public server, they cover core primitives only.

Tool	Description	Parameters
get_primitive_examples	Returns named example scenarios for a primitive	key (string, required), version (string, optional)
scaffold_primitive	Creates a new custom primitive scaffold in the project	name, label, description, category, props
validate_primitive_props	Validates a props object against a primitive's declared contract	key (string, required), version (string, optional), props (object, required)

Guidance

Tool	Description	Parameters
recommend_manifest_structure	Recommends entities, pages, relations from a business goal	goal (string, required)
suggest_page_set_for_entities	Generates CRUD pages for given entities	entities (string[], required)
suggest_relations	Suggests relations from entity definitions	entities (object[], required)
explain_validation_errors	Turns validation errors into fix suggestions	errors (object[], required)

Validation

Tool	Description	Parameters
validate_manifest	Validates a full CellManifestV1	manifest (object, required)
validate_entity	Validates a single EntityDefinition	entity (object, required)
validate_page	Validates a single PageDefinition	page (object, required)
normalize_manifest	Validates and normalizes a manifest	manifest (object, required)

Available resources

The server exposes 5 documentation resources that MCP clients can read for context.

URI	Name	Description
ikary://docs/manifest-format	ikary-manifest-format	Manifest format reference with root structure, spec fields, and semantic rules
ikary://docs/entity-schema	ikary-entity-schema	Entity definition schema with fields and semantic rules
ikary://docs/page-types	ikary-page-types	Page types with available types, fields, and semantic rules
ikary://docs/field-types	ikary-field-types	Field types covering string, text, number, boolean, date, datetime, enum, and object
ikary://docs/primitive-catalog	ikary-primitive-catalog	UI primitive catalog organized by category

Workflow example

A typical AI agent workflow:

1. Call recommend_manifest_structure with a business goal to get a starting structure
2. Build a manifest from the recommendations
3. Call validate_manifest to check for errors
4. If errors exist, call explain_validation_errors to understand what to fix
5. Correct the manifest and validate again
6. Call normalize_manifest to fill in defaults and get the final output

At any point, the agent can call get_manifest_schema or get_entity_definition_schema to look up field definitions, list_primitives to discover UI components, or read a resource for documentation context.