MCP tools

27 token-efficient drop-in replacements for Claude Code's built-in tools, plus 13 ashlr-only utilities.

ashlr ships 40 MCP tools. The bulk of them are drop-in replacements for Claude Code's built-ins — Read, Grep, Edit, Bash, etc. — that compress, summarize, or RAG-route to use fewer tokens. The rest are utilities that don't have built-in equivalents (genome management, savings reporting, GitHub write ops, semantic diffing, etc.).

Every tool's name is prefixed ashlr__ so it never collides with a built-in.

Reads & search


`ashlr__read`	snipCompact head + tail; middle elided. The single biggest token saver.
`ashlr__grep`	Genome-RAG-backed; ~84 % savings on warm queries.
`ashlr__glob`	Globbing with sane head + tail elision on giant result sets.
`ashlr__tree`	Per-dir truncation visible to the model.
`ashlr__ask`	Codebase Q&A backed by the genome.
`ashlr__orient`	One-shot project orientation snapshot.

Edits


`ashlr__edit`	Diff-only edits. The full file is never re-sent.
`ashlr__multi_edit`	Atomic multi-hunk edit.
`ashlr__diff` · `ashlr__diff_semantic`	Read git diffs the model can actually reason about.

Shell & infra


`ashlr__bash`	Auto-summarized stdout for verbose commands.
`ashlr__bash_start` · `ashlr__bash_stop` · `ashlr__bash_tail` · `ashlr__bash_list`	Long-running command lifecycle.
`ashlr__sql`	SQLite + Postgres + MySQL with row-elision.
`ashlr__http`	snipCompact-trimmed JSON/HTML response bodies.
`ashlr__logs`	Tail and grep log files with elision.
`ashlr__webfetch`	URL fetch with markdown + sniff.

Genome


`ashlr__genome`	The genome RAG read path that powers grep.
`ashlr__genome_propose`	Submit a section to the auto-genome loop.
`ashlr__genome_consolidate`	Run the consolidation pass manually.
`ashlr__genome_status`	Inspect the genome cache + cloud-sync state.
`ashlr__savings`	Programmatic access to the per-session ledger.

GitHub


`ashlr__github`	Read-only GitHub helpers.
`ashlr__pr` · `ashlr__issue`	Open/comment on PRs and issues from the model.

Cost-savings methodology

Every tool's per-call savings is measured against the equivalent Claude Code built-in on a fixed corpus of files ≥ 2 KB. See Concepts → Savings math for how the numbers in the per-tool tables are computed.