Integrations
EGC for Grok Build
Give Grok Build the same persistent memory, guardrails and context as 27 other AI coding tools. One local install, encrypted on your machine, nothing sent to a cloud.
Community-supportedInstalled, registered and tested in CI like every other tool; a vendor change that breaks it waits for a report or a pull request. How support levels work.
Install
One command sets up the engine for Grok Build and every other tool it detects. Then run egc init once in each project you want EGC to manage.
npm install -g @egchq/egc && egc install
cd your-project && egc initThe prompt library (agents, skills, commands and rules) is optional and off by default. Add it to every detected tool with egc install --prompt-library, or to Grok Build alone with egc install --target grok --profile full.
What Grok Build gets
Memory
Decisions, session context, working memory and lessons captured in Grok Build are available in every other tool you open, and the other way around. Encrypted at rest, kept per project and branch, never committed.
Session Mesh
Open sessions announce themselves, claim the files they edit and exchange events on a local bus; Grok Build reads the bus through the memory protocol, and through the MCP tools where they are registered.
Guardian
Command validation and write gating by egc-guardian: through its MCP tools where they are registered, and through the tool's own hooks where it has them. Guardian as ~/.grok/hooks/egc-guardian.json, a whole PreToolUse hook file in the Claude Code shape on Bash, Edit and Write plus the native names Grok reports (run_terminal_command, monitor, search_replace and write), denying through grok-guardian-adapter.js with exit code 2 and {"decision": "deny"}; no Token Crusher hook, since Grok's hook docs describe no command rewrite.
Token Crusher
Shell output is compressed before it reaches the model, with every error and warning kept and a local ledger of what it saved. Where the tool has no hook that rewrites commands, the PATH-level shim (egc crusher-shim) covers git, npm, gh and the rest.
Where EGC installs
| Target id | grok |
|---|---|
| Files | Skills installed flat under ~/.grok/skills/<name>/SKILL.md, where Grok discovers them (GROK_HOME moves the whole root); agents, commands and rules ship as library folders under ~/.grok/egc/library/, where nothing of Grok's looks; the root AGENTS.md installs as ~/.grok/AGENTS.md, the global rules file Grok reads before the project ones, where the cognitive bootstrap keeps the memory protocol. |
| MCP servers | Registered by egc init as [mcp_servers.<id>] tables in ~/.grok/config.toml (under GROK_HOME when set), the per-server shape EGC already writes for Codex, keeping every other key in the file. |
The full map of how each tool integrates lives in the integration tiers spec.
Share memory between Grok Build and
Every tool below reads and writes the same memory as Grok Build. Open two of them side by side and the second already knows what the first decided.
