diff --git a/ncp-prototype/README.md b/ncp-prototype/README.md
new file mode 100644
index 0000000..85300f2
--- /dev/null
+++ b/ncp-prototype/README.md
@@ -0,0 +1,14 @@
+## NCP Core Types
+
+This directory contains the Nexus Context Protocol prototype implementation.
+
+### Structure
+
+- `src/types.nim` - Core types (CID, ContextNode, Path)
+- `src/l0_storage.nim` - File backend, CID generation (Blake3)
+- `src/l1_index.nim` - B-Tree index, path-based addressing
+- `tests/test_ncp.nim` - Unit tests
+
+### Status
+
+Feature 1 (Core Types): In Progress
diff --git a/ncp-prototype/src/types.nim b/ncp-prototype/src/types.nim
new file mode 100644
index 0000000..9c340f5
--- /dev/null
+++ b/ncp-prototype/src/types.nim
@@ -0,0 +1,76 @@
+## types.nim: Core Types for Nexus Context Protocol
+## RFC-NCP-001 Implementation
+## Author: Frankie (Silicon Architect)
+
+import std/[tables, options, times]
+
+## Content Identifier (CID) using Blake3
+## 256-bit hash for content-addressed storage
+type CID* = array[32, uint8]
+
+## Content types for Context Nodes
+type ContentType* = enum
+  ctText       ## Plain text content
+  ctImage      ## Image data
+  ctEmbedding  ## Vector embedding (L2)
+  ctToolCall   ## Tool/function call
+  ctMemory     ## Agent memory
+  ctSignature  ## Cryptographic signature
+
+## Context Node: The fundamental unit of NCP
+## Represents any piece of context in the system
+type ContextNode* = object
+  cid*: CID                    ## Content identifier (Blake3 hash)
+  parent*: Option[CID]         ## Previous version (for versioning)
+  path*: string                ## Hierarchical path /agent/task/subtask
+  contentType*: ContentType    ## Type of content
+  data*: seq[byte]             ## Raw content bytes
+  embedding*: Option[seq[float32]]  ## Vector embedding (optional)
+  timestamp*: int64            ## Unix nanoseconds
+  metadata*: Table[string, string]  ## Key-value metadata
+
+## Path utilities for hierarchical addressing
+type Path* = object
+  segments*: seq[string]
+  absolute*: bool
+
+proc initPath*(path: string): Path =
+  ## Parse a path string into segments
+  ## Example: "/agents/frankie/tasks" -> ["agents", "frankie", "tasks"]
+  result.absolute = path.startsWith("/")
+  result.segments = path.split("/").filterIt(it.len > 0)
+
+proc toString*(p: Path): string =
+  ## Convert path back to string
+  result = if p.absolute: "/" else: ""
+  result.add(p.segments.join("/"))
+
+## CID Generation (placeholder - actual Blake3 integration later)
+proc generateCID*(data: openArray[byte]): CID =
+  ## Generate content identifier from data
+  ## TODO: Integrate with actual Blake3 library
+  ## For now: simple XOR-based hash (NOT for production)
+  var result: CID
+  for i in 0..<32:
+    result[i] = 0
+  for i, b in data:
+    result[i mod 32] = result[i mod 32] xor uint8(b)
+  return result
+
+## Context Node Operations
+proc initContextNode*(
+  path: string,
+  contentType: ContentType,
+  data: openArray[byte]
+): ContextNode =
+  ## Initialize a new ContextNode
+  result.path = path
+  result.contentType = contentType
+  result.data = @data
+  result.cid = generateCID(data)
+  result.timestamp = getTime().toUnix() * 1_000_000_000  # nanoseconds
+  result.metadata = initTable[string, string]()
+
+## Export utility functions
+export CID, ContentType, ContextNode, Path
+export initPath, toString, generateCID, initContextNode