When an AI client sends code to be executed to the MCI server instance, MCI
initializes a process. A process in MCI is a stateful, controllable
execution unit.
1. Initialization
A process is created by sending a POST request to /processes with code,
environment configuration, and execution options.
Upon creation, the server validates and optionally lints the process. If
successful, the server assigns a unique Process ID (PID). The process begins
execution unless otherwise specified.
2. Scheduling
If execution is triggered (either at creation or via the startup signal), the
MCI server instance looks to see if the process is dependent on any other
processes and resolves them accordingly, if so, the process enters the queued
state until all it’s dependents are executed.
3. Execution
After resources are free to use and all dependents are executed, the process
enters a running state. During this, the process runs within it’s declared
environment and constraints, logs (stdout, stderr, timestamps) are captured
and structured output may be produced.
It is important to note however that, during this, the proccess can not be
modified but control signals like kill, restart and instantiate can be
sent.
4. Completion
Execution ends when the process transitions to the complete state. Completion
represents the end of runtime activity. It does not imply success. Each
completed process has an exit_state:
success - Execution completed without error.failed - Execution ended due to error.terminated - Execution was externally stopped (e.g., via kill).
The status reflects lifecycle phase, while exit_state reflects outcome of the
process. Last modified on March 14, 2026
