Developer Documentation

Developer documentation for the MO§ES™ AI operator evaluation platform. OpenAPI spec, MCP server with 25 tools, CLI, telemetry schema, and governance constraints. MO§ES™ exposes three integration surfaces for AI agents and developers: an MCP server, an OpenAPI spec, and a CLI.

Run the Demo

The fastest way to understand the platform is to run the 10-step demo pipeline on 50 synthetic operators:

curl -sL https://mos2es.org/demo/run.py | python3 -

Or see the demo page for all run options, including individual commands, the test suite, and the MCP server.

You can also browse the interactive web walkthrough for a visual explanation of each pipeline stage.

OpenAPI Specification

The full API specification is available at /openapi.json. It describes 16 read endpoints and 5 write endpoints across the pilot evaluation surface.

Read tools (16): pilot status, operator profile, cohort distribution, composite score, score summary, diagnostics, data quality, divergence, workflow fit, intervention status, pilot options, configuration validation, operator comparison, executive dashboard, benchmark results, canonical inventory
Write tools (5): assign intervention, close intervention, create experiment, record workflow observation, attach outcome dataset — all require authorization

MCP Server

MO§ES™ ships a Model Context Protocol server with 21 tools. The MCP server allows Claude, ChatGPT, Cursor, and other AI agents to call the platform natively.

Live endpoint: https://mcp.mos2es.org/mcp (Streamable HTTP)

Server info: https://mcp.mos2es.org/

Transport: Streamable HTTP (remote) or stdio (local via Python SDK)

SDK: Python MCP SDK (FastMCP) for local, Cloudflare Worker for remote

Quick start

# Initialize
curl -X POST https://mcp.mos2es.org/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"initialize","params":{},"id":1}'

# List tools
curl -X POST https://mcp.mos2es.org/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","params":{},"id":2}'

# Call a tool
curl -X POST https://mcp.mos2es.org/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"get_pilot_status","arguments":{}},"id":3}'

Tool list

Tool	Type	Description
get_pilot_status	read	Cohort size, observation count, date range, data quality
get_operator_profile	read	Operator metrics, percentiles, benchmarks, divergence flags
get_cohort_distribution	read	Percentile bands for a metric across the cohort
get_composite_score	read	0-100 developmental score (DEVELOPMENTAL, not PERSONNEL)
get_composite_score_summary	read	Aggregate score distribution (no individual rankings)
get_diagnostics	read	Pattern detections and diagnoses (HYPOTHESIS, never fact)
get_data_quality	read	Completeness, coverage, validity metrics
find_usage_operation_divergence	read	Usage vs operation divergence flags
get_workflow_fit	read	Operator-workflow fit analysis
get_intervention_status	read	Active and closed interventions
list_pilot_options	read	Available metrics, eval families, benchmark classes
validate_pilot_configuration	read	Validate a pilot config before deployment
compare_operator_to_reference	read	Compare operator to reference population
get_executive_dashboard	read	Self-contained HTML dashboard
verify_change	read	Verify a measured change after intervention
create_pilot_configuration	read	Generate a pilot configuration from parameters
assign_intervention	write	Assign intervention to operator (requires auth)
close_intervention	write	Close an intervention with outcome notes (requires auth)
create_experiment	write	Create an experiment configuration (requires auth)
record_workflow_observation	write	Record a workflow fit observation (requires auth)
attach_outcome_dataset	write	Attach external outcome dataset (ASSOCIATION, never CAUSATION) (requires auth)

CLI

The enterprise CLI provides 14 command groups for direct access to all platform functionality:

enterprise status
enterprise operator <id>
enterprise cohort --metric leverage
enterprise score composite <operator_id>
enterprise score composite-summary
enterprise benchmark operator <id> --metric leverage
enterprise benchmark cohort --metric leverage
enterprise diagnose <operator_id>
enterprise intervene assign <operator_id> --type <type>
enterprise export dashboard --output dashboard.html
enterprise ingest claude --file <path>
enterprise ingest api-groq --operator op_001 --days 30

Canonical Telemetry

The platform operates on content-free token counts — no prompt text required:

INPUT (I): tokens sent to the AI system
OUTPUT (O): tokens received from the AI system
CACHE READ (R): tokens reused from context cache
CACHE WRITE (W): tokens written to context cache

Metrics

Leverage: (R + W) / I — context reuse and building relative to new input
Yield: O / (I + O + R + W) — productive output share of total token flow
Token SNR: signal-to-noise ratio in token flow
Log Leverage: log-scaled leverage variant
Construction: W / R — ratio of new context built to context reused

Governance Constraints

All composite scores labeled DEVELOPMENTAL, not PERSONNEL
All diagnoses labeled HYPOTHESIS, never fact
All outcome joins labeled ASSOCIATION, never CAUSATION
No bottom-employee leaderboard
No automatic adverse employment actions
No punitive labels
Write operations require authorization

Repository

Public site repo: github.com/SunrisesIllNeverSee/moses (public)

Contact

Email: burnmydays@proton.me