Values ingrained, sovereignty sustained — governance framework for endogenous AI workflows
Every organization has values. The harder question is whether their AI systems actually reflect them. Stated principles and deployed behavior have a way of drifting apart — not through bad intentions, but because translating human values into running software is genuinely hard.
DogmaMCP closes that gap. It's the first integrated framework that brings values-aligned governance to your AI coordination layer — where accountability isn't a promise, but an audit trail.
- EndogenAI is the open source community advancing values-aligned AI governance
- dogma is the framework that encodes your organizational values as rules your AI reads and follows
- DogmaMCP is the experimental tooling that enforces those rules automatically as your AI works
🔒 Local enforcement, not cloud dependencies. Your governance runs on your machine, using files you control. If your network drops, your enforcement layer keeps working.
🌍 EndogenAI | DogmaMCP
A governance framework that embeds organizational constraints into AI workflows — encoding values, principles, and guardrails as executable substrate so agents operate within your constraints by design, not by policy alone.
Governance Layer: MANIFESTO.md • AGENTS.md • Glossary | Quick Start: Get Started
DogmaMCP is experimental. It is actively developed and used in production by its authors, but the public API, MCP tool interfaces, and substrate schema are not yet stable — breaking changes may occur between releases.
What it's ready for: Adoption as a governance template for your own agentic workflows; contributing to the framework; early integration with VS Code Copilot, Claude, and Cursor via the MCP server.
What it's not ready for: Mission-critical production deployments where you cannot absorb breaking changes; integration into regulated/compliance environments without your own review layer.
Where we're headed: Stable 1.0 release targeting Q3 2026 — locking the MCP tool interface, substrate schema, and cookiecutter template contract. See CHANGELOG.md for recent progress.
Questions, early feedback, or integration use cases: open a GitHub Discussion or file an issue.
Q3 2026 → 1.0 Stability Milestone: We're working toward a stable 1.0 release that locks the MCP tool interface, substrate schema, and cookiecutter template contract. This milestone marks the transition from experimental to production-ready, enabling organizations to adopt DogmaMCP with confidence that breaking changes will follow semantic versioning.
Beyond 1.0, we're expanding toolkit coverage to support more AI coordination patterns, improving the contributor onboarding framework to make it easier for new maintainers to join, and refining the governance validation suite based on real-world adoption feedback. Community signal — GitHub stars, forks, discussions, and external adopter reports — directly influences our sequencing.
For the full 12-month strategic roadmap and detailed phase breakdown, see docs/plans/2026-03-10-endogenai-strategic-roadmap.md.
DogmaMCP consists of two interlocking surfaces that reinforce each other:
Permanent Substrate — Encoding layer (MANIFESTO.md, AGENTS.md, scripts/, docs/). Your values and conventions persisted in git, read by every agent before every action. The morphogenetic seed from which all future sessions grow.
MCP Enforcement Layer — Runtime tools (MCP server, validation scripts, pre-commit hooks). Executable machinery that prevents agents from violating encoded constraints. Operates on the permanent substrate; designed to fail fast and surface violations immediately.
Together: substrate + enforcement = governance. Your principles stay in your repo — no external policy servers, no vendor lock-in. Every agent that reads this repository's AGENTS.md and passes substrate validation operates within your governance boundary.
This repository is the reference implementation of dogma, the agentic governance framework maintained by the EndogenAI community. dogmaMCP combines four substrates (Policy Docs, Design/Technical Docs, Agent Files, Enforcement Scripts) into an infrastructural AI harness. Learn more in the glossary.
The dogma MCP server exposes 13 governance tools for session management, validation, research, and scaffolding:
| Tool | Purpose | When to Use |
|---|---|---|
check_substrate |
Full CRD substrate health check | Session start — verify repo is healthy before any phase begins |
validate_agent_file |
Validate .agent.md file against AGENTS.md constraints |
Before committing any agent role file |
validate_synthesis |
Validate D4 research doc before archiving | Before archiving docs/research/*.md files |
scaffold_agent |
Scaffold a new .agent.md stub from template |
Creating a new custom agent role |
scaffold_workplan |
Scaffold a new docs/plans/ workplan |
Multi-phase sessions (≥3 phases or delegations) |
run_research_scout |
Fetch and cache external URL (SSRF-safe) | Research phases; pre-warm source cache before web fetches |
query_docs |
BM25 semantic search over dogma docs corpus | Before fetching external sources; find what's already documented |
prune_scratchpad |
Initialize or inspect session scratchpad | Session start/close; track cross-agent context |
detect_user_interrupt |
Check for user STOP/ABORT/CANCEL signals | Before executing any phase action |
route_inference_request |
Route inference to local or external providers | Prefer local execution (Local-Compute-First) |
normalize_path |
Cross-platform path normalization + env-var expansion | File operations across macOS/Linux/Windows |
resolve_env_path |
Read env-var as path and normalize it | Resolving env-vars like $REPO_ROOT, $HOME to normalized paths |
get_trace_health |
Live OTel trace capture telemetry | Observability; validate telemetry pipeline |
Full documentation: mcp_server/README.md
Visualize governance telemetry, validation state, and MCP tool invocations in a browser dashboard.
Key Features:
- Tool Call Telemetry — Track which tools are invoked, when, and success/failure rates
- Validation Gate Status — Monitor pre-commit hook runs, agent file compliance, synthesis doc quality
- Scratchpad Tracking — View live cross-agent context and session handoff state
- Phase Telemetry — Observe workplan phase progression and inter-phase gates
- Substrate Health — Real-time CRD health check results + last sync timestamp
Try it: uv run --extra web python scripts/start_dashboard.py — opens at http://localhost:5173
Choose your path:
Adopt dogma as a template for your own endogenic governance framework.
# Clone and scaffold
cookiecutter gh:EndogenAI/dogma
# Or: uv run python scripts/adopt_wizard.py
cd <your-project>
uv sync
uv run pytest # verify setupNext: Read docs/guides/getting-started.md for step-by-step guide, customization, and example adoption scenarios.
Fork dogma, add features, and submit PRs back to the EndogenAI project.
git clone https://github.com/EndogenAI/dogma.git
cd dogma
uv sync --extra dev --extra mcp # Install dev tools (pytest, ruff, pre-commit) + MCP server
# Install pre-commit hooks (required before making commits)
uv run pre-commit install
uv run pre-commit install --hook-type pre-push
uv run pytest
# Create feature branch
git checkout -b feat/your-feature
# Make changes...
git commit -m "feat(docs): describe your change" # ← pre-commit hooks run automatically here
git push -u origin feat/your-feature
# Open PR on GitHubWhat gets installed:
| Command | Installs | Use case |
|---|---|---|
uv sync |
Base dependencies only (governance scripts, OTel, docs tooling) | Adopters using dogma as a template; don't need dev tools |
uv sync --extra dev --extra mcp |
Base + pytest, ruff, pre-commit, MCP server | Contributors; running tests, linting, and MCP server locally |
Next: Read CONTRIBUTING.md for conventions and review process.
- GitHub Discussions — Ask questions, share patterns
- Contributing — How to contribute
- Code of Conduct — Community values
- Contact — EndogenAI maintainers: conduct@endogenai.com
"With docs + agent fleet treating docs as dogma → more fundamentally aligned code and content."
The journey: Policy in my head → Tech docs (unanchored) → Agent files (improvised) → Policy docs upstream → Encoded values → Enforcement ✓
AccessiTech is a digital accessibility consultancy built around a single core value: dA11y — my shorthand for digital Accessibility: accessibility as a first-class value, not an afterthought. (Follows the i18n abbreviation pattern: d + 11 letters + y). This is exactly the kind of deep value that dogma exists to encode.
I started here: policy living entirely in my head. Technical documentation helped anchor decisions, but those docs were disconnected from how I actually worked in code — good writing with no enforcement layer. Introducing agent files (AGENTS.md, custom roles, SKILL.md) moved things forward, but early versions were improvised and inconsistent. The key inflection point came when I turned inward: encoding policy documentation upstream, before code, before agent prompts, before PR reviews. That shift made it possible for a value like dA11y to propagate throughout the codebase — not as a comment in a single file, but as a governing constraint that every agent reads before acting.
Getting there required effort: drawing out implicit conventions, researching best practices, iterating on governance docs that had never existed before. But the effort compounded. A comms strategy became a comms policy; a comms policy prompted language-specific accessibility guidelines; guidelines became enumerated rules that enforcement scripts could check automatically.
The result: documentation and an agent fleet that treat those docs as dogma; enforcement scripts, automated tests, and GitHub workflows that catch-and-correct misalignments before they reach production. Overhead became infrastructure. Values became architecture.
Interested in a similar outcome for your codebase? See Get Started.
| Document | Purpose |
|---|---|
| MANIFESTO.md | Core philosophy and three axioms of endogenic development |
| AGENTS.md | Operational constraints for all agents working in this repo |
| CONTRIBUTING.md | Contributor setup, commit discipline, and validation gates |
| docs/guides/ | Formalized workflows (session management, MCP setup, agent authoring) |
| docs/research/ | Research syntheses on governance, AI ethics, and methodology |
| docs/glossary.md | Canonical definitions for all key terms, axioms, and principles |
| .github/agents/README.md | Agent fleet catalog |