How ctx| works
End-to-end flow from repository connection to agent context via MCP.
ctx| sits between your team's source systems and the AI agents that need to understand them. You connect the sources your team relies on, ctx| turns them into searchable context, and agents query that context through MCP when they need it.
The important idea is that ctx| does not ask agents to read everything from scratch. It prepares your code, documentation, and working context ahead of time, then gives agents a controlled way to ask useful questions about it.
Flow
Connect sources
Start by connecting repositories, docs, and other team knowledge sources to a ctx| organization. For GitHub, that usually means installing the GitHub app and choosing the repositories ctx| should be allowed to read.
Each connection is scoped to your organization, so ctx| can keep context, permissions, and agent access aligned with the team that owns the source.
Build useful context
ctx| ingests connected sources so agents can search and reason over them quickly. For repositories, that means cloning the selected ref, building a search index, and extracting structure such as services, APIs, dependencies, libraries, and patterns into the knowledge graph.
Product knowledge matters too. Files like AGENTS.md, skills, ADRs, and
other team instructions help ctx| answer in a way that matches how your
organization works, not just what appears in a single search result.
Give agents one place to ask
Agents connect to ctx| through MCP. Instead of wiring every agent directly to every repo, document store, and internal convention, you give them access to ctx| and let ctx| broker the context they need.
The main tool is ctx_advisor. An agent asks a question, ctx| consults
the organization context it has built, and the agent receives an answer
grounded in your connected sources.
Keep context current
ctx| tracks readiness as sources are indexed and ingested. When connected sources change, ctx| can re-ingest them so agents are not stuck with stale context.
The exact update path depends on the connection type. GitHub repositories, Confluence spaces, and other sources each have their own setup flow, but the goal is the same: keep the context agents use close to the source of truth.
Self-hosting
Self-hosted deployments use the same product model with your own backend, UI, worker, codesearch, Postgres, and graph database. See the self-hosting section for deployment-specific details.