Frigg

Local code context for AI coding agents. Frigg helps Codex, Claude, Cursor, and other MCP clients work from concrete repository evidence instead of broad scans, whole-file dumps, or model-memory guesses.

Quickstart Configure clients LLM Context

Frigg is for developers who use coding agents on real repositories and want those agents to inspect code the same way a careful engineer would: find the right files, follow the relevant symbols, read bounded source windows, and point back to evidence.

It turns each repository into local code context for agent work. A shared MCP service gives clients and subagents one vocabulary for source-backed search, navigation, and reads without making the prompt carry the whole repository.

Why Frigg exists

Generation is only half of agentic coding. The right answer starts with the right repository evidence.

Frigg is useful when a shell scan, a full-file dump, or a best-guess answer is no longer enough. It gives agents a local map of the codebase so they can discover relevant files, narrow the context, and answer from source instead of memory.

What changes

Agents spend less time guessing which files matter.
Source reads stay small enough to fit the task.
Humans can inspect the same paths and source windows the agent used.
Teams can standardize how Codex, Claude, Cursor, and subagents inspect a repository.

How Frigg fits

Frigg runs locally. You install one binary, initialize repository state under .frigg/, index the source, and connect MCP clients to the shared Frigg service.

The default path is intentionally direct: local repository evidence first, with optional semantic recall and optional SCIP-backed precision when a workflow needs them.

Start here

Quickstart gets one repository connected to an MCP client.
Installation lists supported install paths.
Client setup explains shared HTTP and single-client stdio modes.
Project adoption installs managed agent instructions, MCP entries, and optional skills.
Agent workflow maps common code-inspection tasks to Frigg tools.

Boundaries

Frigg is not an AI pair programmer, IDE replacement, hosted code intelligence platform, Copilot alternative, or generic semantic search product.

Its narrow job is local code evidence for AI agents. Normal indexing does not edit project source files; Frigg updates its own .frigg/ state. Optional cloud or self-hosted semantic providers can receive source-derived embedding inputs, and optional precise generators can execute tools and write .frigg/scip/ artifacts.

Go deeper

Source Anchors

../frigg/README.md
../frigg/Cargo.toml
../frigg/crates/cli/src/mcp/guidance.rs