Skip to content

For Claude Code users

Claude starts every session blind.

CLAUDE.md tells Claude how to behave. It does not tell Claude what's in your repo. Maguyva indexes the codebase and serves it back over MCP, so Claude edits like it has already read the code.

Index your first repo Read the Claude Code setup

Public beta. Free tier: 3 repos, 50K indexed lines, no card.

Use CLAUDE.md for behavior. Use Maguyva for current codebase facts.

CLAUDE.md is a best practice. Use it.

> cat CLAUDE.md
# project conventions, commands, scoped rules

Persistent instructions belong in CLAUDE.md. It is the right place for:

  • Build, test, and lint commands the agent should run.
  • Coding-style rules and naming conventions.
  • “Always do X / never do Y” guardrails scoped to a directory.
  • Pointers to the canonical docs and decision logs.

Anthropic’s memory docs describe this well. Keep it concise. Scope it. Commit it.

But CLAUDE.md was never meant to be a map of your codebase.

Where CLAUDE.md stops being enough

Four failure modes that markdown alone can’t fix.

// the file isn’t the codebase

Telling Claude how to behave doesn’t tell it what exists. The first edit on an unfamiliar repo is a guess about file paths, function names, and call sites. CLAUDE.md cannot list every symbol, and you wouldn’t want it to.

// docs go stale, code does not

A CLAUDE.md block describing your auth flow is right until someone refactors auth. The code is now the source of truth and the docs are confidently wrong. Claude reads the wrong one.

// dependencies aren’t in markdown

“What breaks if I change this function?” is a graph question. It cannot be answered by a doc file. Claude either grep-and-prays or asks you to paste files into the chat.

// context windows have a budget

Stuffing CLAUDE.md until Claude “knows enough” eats tokens that should pay for reasoning. Beyond a few KB you trade answer quality for context volume.

Maguyva is the layer CLAUDE.md doesn’t cover

A remote MCP server that gives Claude Code:

  • Semantic, AST, graph, and text search over your indexed repo, returned with file paths and line numbers.
  • Dependency, type, and call-site lookups the questions a graph answers and a doc file cannot.
  • Branch-aware results so Claude sees the version of the code it’s actually editing.
  • Workspace pricing not per-seat. Index 1 repo or 50; agents don’t pay seats.

CLAUDE.md tells Claude how to behave.

Maguyva gives Claude a queryable map of the code.

Three workflows

Numbered, code-block-heavy. The questions you already ask Claude, grounded against real lines.

// workflow 01

Find the real implementation, not the test stub

You: "How does retry work in our payment client?"

Without Maguyva → Claude greps retry, finds 14 hits, picks one (often a test mock).
With Maguyva    → Maguyva returns the symbol definition, the call-sites, and the
                   file:line of the real implementation, ranked.

// workflow 02

Check blast radius before you refactor

You: "What calls normalizePhoneNumber?"

Maguyva returns: 7 call-sites across 4 packages, including one in a Python service
that imports it via the gRPC stub. Claude proposes the change with the migration
list inline, not after CI goes red.

// workflow 03

Verify the agent before you accept the diff

Claude: "I edited src/auth/session.ts:142 to fix the token refresh."

Ask Maguyva: "Show me session.ts:130-160 and anything that imports session."

Maguyva returns the live file slice + 3 importers. The diff is now grounded
against real lines, not Claude’s recollection from token #11,000.

Setup in Claude Code

Three steps. Free tier: 3 repos, 50K indexed lines, no card.

  1. // step 01

    Index a repo at maguyva.ai

    Pick one you know well so you can verify the answers. Free tier covers 3 repos, 50K indexed lines.

  2. // step 02

    Add Maguyva to Claude Code as a remote MCP server

    $ export MAGUYVA_API_KEY=mgv_xxxx
$ claude mcp add --transport http maguyva https://maguyva.tools/mcp \
    --header "Authorization: Bearer ${MAGUYVA_API_KEY}"

  3. // step 03

    Ask one question whose answer you already know

    Start with one repo and one verifiable question, not your whole company. If the answer matches what you’d say, you’re set.

Full instructions live in Maguyva’s docs.

Public beta. Self-serve signups open.

Index one repo you already know.

Test one question whose answer you can verify. That’s the whole onboarding.

Get a beta spot See compatibility docs

free plan: 3 repos, 50K indexed lines, no card.