cordis-bridge 0.1.2

CLI and optional MCP bridge for the Cordis cognitive runtime.

Cordis Bridge v0.1

cordis-bridge exposes the existing CORDIS cognitive loop to a real host. It has no model, planner, HTTP server, or execution permission of its own.

The bridge provides one shared Service behind two transports:

• a JSON-only CLI for local hosts and scripts;
• an optional stdio MCP server for agent hosts.

host -> begin -> compact context + control signal
host -> observe / check-action -> workflow evidence
host -> finish -> Core feedback + Memory episode
later host -> begin -> changed cognition

By default, durable state lives in .cordis/ below the current directory:

.cordis/state.json      Core learning state
.cordis/cognition.db    Memory and relationship graph
.cordis/focus.json      Active host-task focus for CLI/MCP process recovery

CLI

Every command reads one JSON object from a file or standard input and emits one JSON object. The input file defaults to - (standard input).

cordis begin task.json
cordis query query.json
cordis observe event.json
cordis check-action action.json
cordis finish feedback.json
cordis status

Use --data-dir PATH before the command to keep a project-specific state directory elsewhere.

MCP

Install the optional transport dependency and run:

pip install "cordis-bridge[mcp]"
cordis-mcp --data-dir .cordis

The server exposes cordis_begin, cordis_query, cordis_observe, cordis_check_action, cordis_finish, and cordis_status. Each tool takes the same JSON contract as the matching CLI command.

Concurrency boundary

CORDIS v0.1 assumes one mutating host per state directory. Run one MCP server per .cordis directory and keep CLI mutations for that directory sequential. The Core JSON state and Bridge focus snapshot are deliberately small local stores; multi-writer coordination is not part of this release.