Context packs
Agents request compact or detailed project context: profile, recent activity, active work, notes, resolved rules, known issues, policies and document summaries.
Operational context for AI coding agents
Project context for AI coding agents.
It connects Copilot, Claude, Cursor and other MCP clients to a governed knowledge base: project context, resolved rules, package policies, workflows, open coordination and audit in one compact response.
- Token-aware context packs
- Workflow-guided MCP tools
- Private by deployment choice
Connects operational knowledge to
GitHub Copilot
Claude Code
Cursor
VS Code Agent Mode
Visual Studio
Any MCP client
AI agents move fast. ATLAS makes the surrounding system visible.
Without ATLAS
- ! House rules live in prompts, chats and review comments.
- ! Agents miss current blockers, active work and recent decisions.
- ! Package and stack choices are corrected late in pull requests.
- ! External MCP tools run without your company context.
- ! Knowledge quality, workflow coverage and audit gaps stay hidden.
With ATLAS
- ✓ Agents fetch a summary-first project context pack before acting.
- ✓ Work items, notes, decisions and documents travel with the project.
- ✓ Rules resolve by global, stack, group and project scope.
- ✓ External tools are routed, enriched and audited through ATLAS.
- ✓ Health Center findings show stale content, broken references and project risks.
A practical control layer for agentic development.
Scoped governance
Guidelines, package policies and known issues resolve across global, stack, group and project scopes, including overrides and stricter project rules.
Review lifecycle
Draft, InReview, Published, Deprecated and Superseded states keep agent-visible guidance intentional while preserving history.
Health Center
Knowledge and project findings surface stale reviews, broken references, duplicate guidance, missing metadata, token pressure and operational risks.
Human and agent coordination
Members, project memberships, work items, notes, blockers and decisions give parallel agents a shared view of who is doing what.
Durable project memory
Changelog entries and project Markdown documents preserve plans, reviews, runbooks and handoffs so the next agent does not start cold.
Workflow recipes
Feature, bugfix, refactor, package-change and review workflows carry phase guidance, tool bundles, guardrails and health checks.
External MCP gateway
Upstream MCP tools are exposed with ext_ prefixes, enriched with project context, governed by policy and captured in the audit trail.
Private operations
Run ATLAS with Docker, nginx, PostgreSQL and pgvector. Use local Ollama embeddings or OpenAI/Azure OpenAI when that fits your boundary.
The loop is simple: publish context, guide work, record what changed.
-
01
Model your environment
Create projects, stacks, groups, members and memberships. Import existing knowledge through the Atlas format when useful.
-
02
Govern the knowledge
Reviewers publish guidelines, package policies, known issues and documents. Drafts stay out of default agent context until they are ready.
-
03
Agents work from ATLAS
A workflow-aware agent calls
get_project_context, checks compliance, looks up packages, claims work and asks focused questions through MCP tools. -
04
The system learns safely
Changelog, notes, documents, reports, health findings and audit entries make the next session better without relying on memory or copy-paste.
5
lifecycle states from Draft to Superseded
4
scope layers: global, stack, group, project
50+
MCP-facing operations for context and governance
1
activity feed across notes, documents, work and changelog
Short answers for technical teams.
Do we have to replace our current coding assistants?
No. ATLAS speaks MCP, so it can sit behind the assistants your team already uses. The assistants keep their UI; ATLAS supplies the context and guardrails.
Is this a wiki, a prompt library, or a policy engine?
It borrows a little from each, but the important difference is delivery. ATLAS resolves the right knowledge for a project and hands it to the agent at work time, with lifecycle, scope and audit attached.
How do agents avoid stepping on each other?
They can claim work items, read unresolved questions and blockers, post high-signal notes, link workflow IDs, and release the work when done. Humans see the same coordination state in the Management App.
Can external MCP tools be used safely?
Yes. ATLAS can proxy upstream tools with ext_ names, enrich calls with project context, route them through policy and record audit entries.
Does code or knowledge have to leave our network?
No. ATLAS can run on your infrastructure with PostgreSQL and pgvector. Embeddings can be generated locally through Ollama, or through OpenAI/Azure OpenAI if your policy allows it.
Where should a team start?
Start with one active repository, its stack rules, a few package policies, known issues and current work items. The first useful context pack usually appears before the knowledge base feels complete.
Bring your agent workflow under control
See ATLAS against a real repository.
Book a short walkthrough. We can show the Management App, MCP server, context packs, workflows, health findings and audit trail on a realistic codebase.