cellflow

Cell+Flow protocol as a pure Python library. Zero MCP dependency — no server to start, no MCP host required. cellflow-mcp is the MCP transport skin on top of this library (SPEC §10: transport-agnostic — protocol capability = library; MCP is just one skin).

pip install cellflow

from cellflow import CellFlow

cf = CellFlow(storage_path="./my_data")           # local SQLite, no server
cell = cf.cell_register(schema_type="agent.v1", genome={"name": "my agent"})
cf.flow_capture(topic="test", content="hello")
print(cf.calculate_if(cell["cell_id"]))           # IF: earned trust, 0-100

Layout (mirrors the protocol)

Deeper internals stay importable at cellflow.modules.* (e.g. cellflow.modules.karma) — same modules, one physical copy.

Injectable storage (P2)

Default is local SQLite. Pass any object implementing the Storage duck-type (read / write / append / delete / search / upsert):

cf = CellFlow(storage=MyCustomStorage())

Or migrate existing SPEC-shaped data losslessly:

cf.cell_import({"cells": [...], "flows": [...]})   # SPEC §1/§2 field layout

Relationship to cellflow-mcp

cellflow-mcp depends on this package and only adds the 56 @mcp.tool() pass-throughs, the CLI, and the HTTP facade. All protocol behaviour — IF/Karma math, hash chains, discovery ranking, bridges — lives here and is identical through every transport.

MIT License · Python 3.10+ · SPEC is the authoritative source.