Skip to content

Integrations

CodeCortex exposes one agent-facing MCP surface and includes merge-safe project configuration for supported coding agents.

Supported coding-agent targets

The current implementation supports these project-local targets:

Agent CodeCortex-managed configuration
Claude Code .mcp.json
Codex .codex/config.toml
Cursor .cursor/mcp.json
Gemini CLI .gemini/settings.json
OpenCode opencode.json

These targets come directly from codecortex.integrations.agents.AgentTarget.

Detect local agents

cortex agents detect

Detection checks supported executables and existing project-local configuration directories/files.

Preview configuration

Before changing any project configuration:

cortex agents configure --dry-run

To preview all supported targets:

cortex agents configure --all --dry-run

Configure detected agents

cortex agents configure

Or configure every supported target:

cortex agents configure --all

The configurator merges the CodeCortex MCP entry into existing JSON/TOML configuration instead of replacing the whole file. Existing files are backed up before a real modification where applicable.

MCP command

All supported project-local integrations ultimately point to the same CodeCortex command:

cortex mcp --path /absolute/path/to/repository

The stdio MCP server exposes repository intelligence through a stable surface while internal engines remain behind the CodeCortex gateway.

Why the integration boundary matters

A coding-agent host should not need to know how repository indexing, symbol resolution, retrieval, Git intelligence, memory, architecture inference, or impact analysis is implemented.

CodeCortex keeps that orchestration behind a single gateway so:

  • the agent configuration stays small;
  • evidence providers can evolve without rewriting host integrations;
  • local-first behavior remains explicit;
  • unavailable optional providers can degrade cleanly instead of pretending to succeed;
  • guarded editing and validation remain inside the same trust boundary.

Remote operation

For shared or hosted deployments, see Distributed operation and the Platform overview. Remote deployments should use the repository's authentication, TLS, policy, quota, and audit guidance instead of exposing internal worker services directly.