Federated AI Commons MCP server

by Toby-Self·io.github.Toby-Self/federated-ai-commons·v0.1.1

No reviews yet

MCP server for a federated AI-commons governance simulation with a verified compliance oracle.

A85/100grade A

Write a review Install

What users say

No reviews yet

Be the first

Safety scan

A85/100

full report

Adoption

New

0 stars27 downloads/wk

Reviews

Write one

Nobody has reviewed Federated AI Commons yet.

If you have run it, two minutes of your experience saves the next person an afternoon.

Federated AI Commons tools (9, 4 write)

write = sends, deletes, buys or posts

Read from the package source without running it. The installed server may list more.

check_governance_compliance
Stateless compliance check: given a proposed action and the state needed to judge it (both supplied in action_json, nothing retained server-side between calls), returns whether the action is compliant with this framework's governance rules -- and why.
compare_configurationswrite action
Run two model configurations across multiple seeds and return their averaged outcome metrics side by side. This is the actual pattern used throughout this project's development -- never trust a single seed; always compare configurations against each other directly rather than reasoning about one in isolation. Several real bugs in this codebase were only caught by exactly this kind of paired compar
get_findings
Return the full contents of FINDINGS.md -- the project's experimental record, including every bug found and how, and current known open issues. Read this before running new experiments to avoid re-discovering an already-established result.
get_readme
Return the full contents of README.md -- the project's quick-start map and condensed parameter reference.
list_compliance_rules
List every available governance compliance rule, its required action_json schema, and the rationale behind it -- read this before calling check_governance_compliance() to construct valid requests.
list_parameters
List every opt-in model parameter with its default value, and separately list the 28 independent '_enabled' master switches -- the top-level subsystems worth knowing about before diving into the full ~132-parameter surface.
run_simulationwrite action
Run a single Federated AI-Commons Model simulation and return key outcome metrics as JSON.
run_test_suitewrite action
Run the persisted pytest regression suite and return a pass/fail summary. Run this before trusting results from a modified copy of the model file -- confirms the ~13 load-bearing findings this whole project depends on still hold, the same check that caught two real problems (a false regression alarm, an actual reversed finding) the one time this suite was itself first tested.
wide_seed_sweepwrite action
Run a configuration across many seeds specifically checking for crashes, assertion failures, or resource blowups -- the exact adversarial-testing pattern that caught every real bug found in this project's development (a billions-scale payout blowup that only appeared in 3 of 10 seeds, a negotiation over-commitment bug, several calibration failures). Returns a report of any seed that failed, not ju

Public scan report

scanner v0.1.9 · 2026-09-25 · same rubric, same numbers if you re-run it

no findings

Code scan9 source files scanned25/25
–Live reliabilityno gateway calls yet and no remote to proben/a
–Tool poisoningtools not inspected (local package is not executed); not countedn/a
Auth qualitylocal package, no credentials required12/15
Maintenancelast push 73 days ago12/15
Maintainer identityregistry namespace matches repository owner6/10

Overall 85/100. Components that don't apply are left out of the denominator. Any critical finding is an F.Rubric Appeal a finding JSON

How Federated AI Commons compares

Against every graded MCP server, from our own scans and uptime checks. Updated after each scan run.

Safety score: 85/100; higher than 70% of 48,255 servers
Tools: 9; the typical server has 7
Tools that change things: 44%; typical: 0%. Fewer write tools means less an agent can break
Downloads a week: 27; more than 14% of 12,191 servers with a package

What its tools let an agent do: 5 read · 2 change · 2 run or deploy

Install Federated AI Commons in Claude Code, Cursor or VS Code

claude mcp add federated-ai-commons -- uvx federated-ai-commons-mcp

Add to Cursor

What the publisher says

From the Federated AI Commons repository's README, as published. We do not edit it. Read it on GitHub

Asymptotic Ethics Model

An agent-based simulation testing whether a post-scarcity governance framework — built on Elinor Ostrom's commons-governance principles — actually holds up when implemented and adversarially stress-tested, rather than just argued for in theory. Includes a working MCP server that exposes the simulation as callable tools for AI agents, and a stateless "governance primitive" that lets an agent check whether a proposed action complies with the framework's rules, verified against the simulation's own logic rather than reimplemented from a description of it.

License: Apache 2.0. Core dependency: Mesa (agent-based modeling in Python).

What makes this worth a look

Most of what's genuinely worth reading here isn't the simulation's existence — it's the discipline behind it. Every mechanism was built expecting to find its own failure mode, and several real bugs, exploits, and false conclusions were caught and fixed as a direct result, not hidden after the fact:

permanent crisis rule under default conditions (~70–90% of the time), not occasional intervention — a concrete demonstration of a real AI governance risk. It was redesigned around consent (a citizen vote) rather than force, and the same pattern was later reused for ecological limits and shared-resource conflicts.

A hard-threshold emergency-governance mechanic was found to produce

vulnerable to capture: free-riding strategies, forming a numerical majority, voted themselves into the most exploitable governance policy — a real result, not a hypothetical concern about democracy.

Democratic policy voting was tested directly and found genuinely

unearned resources into status before being fixed; a barter-economy value function was found to make trading a strictly worse strategy than isolation before being fixed; an evolutionary fitness metric was found to collapse an entire economy into monoculture before being fixed.

A reputation-reward mechanism was found to let free-riders launder

the simulation's real decision logic and verified by capturing actual votes and transfers from a live run — not just checked against themselves, which would prove nothing.

The "governance primitive" compliance rules are extracted verbatim from

The full record of every finding, bug, and fix — in the order it happened — is in FINDINGS.md.

Quick start

pip install mesa pytest mcp networkx numpy

from asymptotic_ethics_model import AsymptoticEthicsModel

m = AsymptoticEthicsModel(200, 8, seed=1, coupled_governance=True,
                             rehabilitation_enabled=True, graduation_enabled=True)
for _ in range(300):
    m.step()

df = m.datacollector.get_model_vars_dataframe()
df.tail(10)

Run the regression suite before trusting any change:

pytest test_asymptotic_ethics_model.py -v

To run the MCP server (exposes the simulation as tools for an MCP client like Claude Desktop — see the header comment in asymptoticethicsmcp_server.py for exact client configuration):

python3 asymptotic_ethics_mcp_server.py

What's in this repo

Shortened. The full README is on GitHub.

Nothing above is checked by us. What we check is on the safety report.

Federated AI Commons: common questions

Is Federated AI Commons MCP server safe?: Yes, by our scan: it is graded A (85/100). Read the Federated AI Commons safety report
How do I install Federated AI Commons?: It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.
Does Federated AI Commons need an API key?: Not as far as the registry entry and our scan can tell: no credentials are declared or required.
Is Federated AI Commons maintained?: The last commit was 113 days ago (2026-07-14). The latest release is v0.1.1.
How does Federated AI Commons compare with other MCP servers?: safety score 85/100 (higher than 70% of 48,255 servers); tools 9 (the typical server has 7); tools that change things 44% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.