Contexity exposes tools through MCP (Model Context Protocol) so agent hosts can retrieve and write context without direct access to the local store. These tools are called automatically by hosts like Codex and Claude Code — you do not call them manually during normal work.
These tools manage the lifecycle of a single agent run: starting it, recording events during execution, and closing it out with metrics and checkpoint data.
Project Context Tools
These tools give the agent access to project context: attaching a project, bootstrapping a fresh session, retrieving a bounded context pack, and recording or invalidating facts.
These tools expose the project graph so the agent can reason about structure, search for relevant code surfaces, and map the impact of changes before and after edits.
External Context Tools
These tools let the agent capture and manage external references — Slack threads, issue links, documentation pages, or inspiration sources — and the signals extracted from them.
These tools manage global preferences that persist across all projects on the machine: reading verified preferences, storing explicit ones, removing outdated ones, and proposing uncertain candidates for later verification.
These tools let the host read and update the metrics display policy, controlling whether post-run usage summaries are shown to you in the terminal.
MCP tools are intended for hosts. During normal work, you should not need to call them manually.
