Migration from Python agemcp

age-mcp is a drop-in replacement for the Python agemcp. Existing data works without migration.

What’s Compatible

Aspect	Python agemcp	F# age-mcp
Tenant prefix	t_{TENANT_ID}__	t_{TENANT_ID}__
Vertex identifier	ident property	ident property
Edge start	start_ident property	start_ident property
Edge end	end_ident property	end_ident property
Edge identifier	ident property	ident property
Properties	Arbitrary JSON in agtype	Arbitrary JSON in agtype
Embeddings table	vertex_embeddings	vertex_embeddings
Vector dimensions	384 (BAAI/bge-small-en)	Configurable (default 384)

Switching

1. Stop the Python agemcp server
2. Point age-mcp at the same database:

# Python used:
# DB__DSN=postgresql+asyncpg://user:pass@localhost:5435/agemcp

# F# equivalent:
AGE_CONNECTION_STRING="Host=localhost;Port=5435;Database=agemcp;Username=user;Password=pass"
TENANT_ID="default"

1. Update your MCP client config:

{
  "mcpServers": {
    "age-mcp": {
      "type": "stdio",
      "command": "age-mcp",
      "env": {
        "AGE_CONNECTION_STRING": "Host=localhost;Port=5435;Database=agemcp;Username=user;Password=pass",
        "TENANT_ID": "default"
      }
    }
  }
}

1. Verify: ask your assistant to list_graphs – you should see all your existing graphs.

Tool Mapping

Python agemcp	F# age-mcp	Notes
get_or_create_graph	get_or_create_graph	Same
list_graphs	list_graphs	Same
drop_graphs	drop_graphs	Same
upsert_vertex	upsert_vertex	Same
upsert_edge	upsert_edge	Same
upsert_graph	upsert_graph	Same
drop_vertex	drop_vertex	Same
drop_edge	drop_edge	Same
cypher_query	cypher_query	Same
search_vertices	search_vertices	Same
search_edges	search_edges	Same
get_neighbors	get_neighbors	Same
export_graph	export_graph	Same
import_graph	import_graph	Same
generate_visualization	generate_visualization	Same
semantic_search	semantic_search	Needs EMBEDDING_API_URL
graph_context	graph_context	Needs EMBEDDING_API_URL
sync_to_openbrain	sync_to_openbrain	Same
import_from_openbrain	import_from_openbrain	Same
–	cypher_write	New: dedicated write tool
–	get_schema	New: label counts

Embedding Differences

The Python version uses fastembed (local ONNX model, BAAI/bge-small-en-v1.5, 384 dimensions). The F# version uses an HTTP embedding API (OpenAI-compatible).

If you have existing embeddings in the vertex_embeddings table from the Python version, the F# version can query them directly. For new vertices, you need to configure EMBEDDING_API_URL.

To use the same model dimensions:

EMBEDDING_API_URL="http://localhost:11434/v1/embeddings"  # Ollama
EMBEDDING_MODEL="bge-small-en-v1.5"
EMBEDDING_DIMENSIONS=384

What’s Not Supported

• SSE / HTTP transport – age-mcp supports stdio only. For Claude Desktop and Claude Code, this is the standard transport.
• Configurable ident property names – Python supports AGE__IDENT_PROPERTY etc. age-mcp hardcodes ident, start_ident, end_ident (the defaults).