Multi-Agent Support

Code Context Engine works with any AI coding agent that supports MCP (Model Context Protocol). The cce init command auto-detects which agents are present in your environment and configures them automatically.

The --agent flag

cce init                   # Default. Detects installed agents.
cce init --agent claude    # Configure only Claude Code
cce init --agent copilot   # Configure only VS Code / Copilot
cce init --agent codex     # Configure only Codex CLI
cce init --agent pi        # Configure only Pi
cce init --agent all       # Configure all supported agents

When no --agent flag is provided, cce init defaults to auto, which scans for known config files and editor directories.

Supported Editors and Agents

Agent	MCP Config	Instruction File	Scope	Detection
Claude Code	.mcp.json	CLAUDE.md	Project	Always configured
VS Code / Copilot	.vscode/mcp.json	.github/copilot-instructions.md	Project	.vscode/ exists
Cursor	.cursor/mcp.json	.cursorrules	Project	.cursor/ or .cursorrules exists
Gemini CLI	.gemini/settings.json	GEMINI.md	Project	.gemini/ or GEMINI.md exists
Codex CLI	~/.codex/config.toml	AGENTS.md	User (global)	~/.codex/ or VS Code OpenAI extension
OpenCode	opencode.json	(none)	Project	opencode.json exists
Tabnine	.tabnine/agent/settings.json	TABNINE.md	Project	.tabnine/ exists
Pi	.mcp.json	AGENTS.md	Project	.pi/ exists

How it works

Each agent integration does two things:

1. Registers the MCP server so the agent can call context_search and other CCE tools.
2. Writes an instruction file telling the agent to prefer CCE’s search over raw file reads.

The instruction file content is managed by CCE and wrapped in markers (CCE:BEGIN / CCE:END) so it can be updated on upgrade without touching your own content.

Cross-agent memory

Decisions, code areas, and session history are stored per-project in memory.db, not per-agent. If you switch between Claude Code and Codex on the same project, session_recall returns decisions from all prior sessions regardless of which agent created them.

Re-running for additional agents

You can run cce init --agent <name> multiple times. Each run is additive and will not remove previously configured agents.

cce init --agent claude
cce init --agent copilot   # Adds Copilot config alongside Claude

Or configure everything at once:

cce init --agent all

Common issues across all agents

”cce: command not found”

The cce binary must be on your PATH. Default locations by install method:

Install method	Binary location
uv tool install	~/.local/bin/cce
pipx install	~/.local/bin/cce
pip install	Depends on your Python environment

Add ~/.local/bin to your shell profile (~/.zshrc, ~/.bashrc, or equivalent).

Agent doesn’t use context_search

1. Check the instruction file exists (CLAUDE.md, AGENTS.md, .cursorrules, etc.)
2. Verify it contains the ## Context Engine (CCE) section
3. Restart the agent after setup
4. Re-run cce init if the instruction file is missing

Savings not updating

Savings only increment when the agent calls context_search or expand_chunk. If the agent uses Read/Grep directly, no savings are recorded. Check cce savings for a “last query” timestamp to confirm whether new queries are happening.

Windows encoding errors

Upgrade to CCE v0.4.24+ which adds explicit UTF-8 encoding to all file I/O. Earlier versions can crash with UnicodeDecodeError when config files contain non-ASCII bytes.