jiminy
Serves a versioned constitution (mission and principles) over MCP as a coherence control plane for multi-agent systems, enabling agents to fetch the current direction and receive change notifications.
README
Jiminy
A coherence control plane for multi-agent systems: one versioned source of truth for your system's mission and principles (its constitution), served over MCP so every agent can act on the direction in force right now — even when that direction changes mid-flight.
The name is the point: a small conscience the rest of the system consults.
Why
When a multi-agent system's goals change, agents holding a stale copy of the old direction keep producing work for it — and worse, quality checks against the stale copy actively push work back toward the obsolete goal. Jiminy removes the stale copy: direction lives in one versioned store, agents pull the current version at each step boundary, and subscribers are notified the moment it changes.
Quick start
pip install . # from a clone; CLI: jiminy
jiminy init-db
jiminy set --mission "Ship a lending product people trust" \
--note "initial constitution"
jiminy current
jiminy serve --transport stdio # or --transport http
A constitution is a mission (the overarching purpose — the tie-breaker when principles conflict) plus ordered principles. Every change appends a new immutable version with a plain-language change note; nothing is edited in place, so "what was the direction when agent X acted" is always answerable.
The contract
Three operations, over MCP or Python:
get_constitution— the direction in force now (mission, principles, version).get_changes_since(known_version)— the pull an agent makes before a step: the full current direction plus the change notes since the version it last saw. A missed notification is harmless — the next pull is self-describing.set_direction(mission?, principles?, change_note)— append the next version. Omitted fields carry forward. On HTTP this requires the bearer token.
Clients may also subscribe to the jiminy://constitution/current resource and
receive a standard MCP resources/updated notification on every version bump.
Multiple constitutions
One Jiminy can hold several constitutions side by side — say one per product
line or per jurisdiction. Every operation takes an optional constitution
name, over MCP and on the CLI (--constitution eu), and defaults to
"default", so a single-constitution setup never has to mention it. Each name
has its own version sequence: bumping eu to v2 leaves default at whatever
version it was. A name you have never written to reads as the same version-0
empty state an untouched store does. The subscribable resource is the default
constitution's; agents on another one pull it by name with get_changes_since.
Storage
SQLite out of the box; PostgreSQL for production via JIMINY_DATABASE_URL.
Storage is pluggable: hand SqlConstitutionStore your own SQLAlchemy Engine
to live inside an existing database, or implement the small store protocol to
bring your own persistence entirely. Concurrent writers are safe — versions are
serialized by a unique index and a retry, never lost or duplicated.
Reads never fail on an empty store: before any direction is set, consumers get a version-0 empty state, so integrating Jiminy ahead of adopting it costs nothing.
Auth
- stdio: open. A process that can spawn the server already owns the database file under it; a token there would be ceremony, not a boundary.
- HTTP: a shared bearer token (
JIMINY_TOKEN) gates every request. The server refuses to start tokenless over HTTP unless you explicitly opt in (JIMINY_ALLOW_INSECURE_HTTP, local experimentation only — it warns).
Testing
python -m pytest -q # SQLite, no network
JIMINY_TEST_POSTGRES_URL=postgresql+psycopg://… python -m pytest -q # + Postgres
Sibling project: Canon tests whether your system's outputs actually cohere with the constitution Jiminy serves.
See CONTRIBUTING.md for style and test expectations.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.