docmancer 0.4.6

Compress local documentation context for coding agents.

docmancer

Compress documentation context so coding agents spend tokens on code, not docs.

Get Started | What It Does | Supported Agents | Docs

---

Docmancer fetches documentation, normalizes it into inspectable sections, indexes those sections with SQLite FTS5, and returns compact context packs with source attribution. The goal is agentic runway: your agent should burn tokens on implementation, tests, and debugging, not on rereading entire documentation sites.

Product shape: an MIT-licensed CLI on PyPI. You point it at a docs URL or local path with add, it indexes sections into a local SQLite database, and your coding agent calls docmancer query through an installed skill. There is no hosted query API, no servers, and no API keys on the core path.

In a typical agentic coding session, raw docs pages can consume a sizable chunk of the context window. Docmancer's context packs are usually a small fraction of the original page weight (each query reports the actual savings on that call), so the agent stays sharp longer and produces more output per session.

---

Quickstart

pipx install docmancer    # Python 3.11, 3.12, or 3.13

docmancer setup
docmancer add https://docs.pytest.org
docmancer query "How do I use fixtures?"

setup creates ~/.docmancer/docmancer.yaml, initializes ~/.docmancer/docmancer.db, and installs detected agent skills. Use setup --all for non-interactive installation across all supported agents. If pipx picks an unsupported interpreter, pin one explicitly: pipx install docmancer --python python3.13.

---

What It Does

Two complementary surfaces, both running entirely on your machine:

Docs RAG (the original): fetch documentation, index normalized sections with SQLite FTS5, return compact context packs with source attribution. Your agent burns tokens on implementation, not on rereading docs.

API MCP packs (runtime support landing): install version-pinned MCP packs compiled from public OpenAPI, GraphQL, TypeDoc, and Sphinx sources, then expose them to your agent through one shared docmancer mcp serve process. Today's runtime is uneven by source kind:

• OpenAPI 3.0 / 3.1: live wire calls via the http executor. Auth is gated, destructive calls are gated behind explicit opt-in, idempotency keys are auto-injected and reused on retry, version pins are enforced on the wire (e.g. Stripe-Version: 2026-02-25.clover).
• GraphQL: the pipeline compiles introspection JSON into operations; a dedicated GraphQL executor is not yet wired in the runtime, so calls return documentation only.
• TypeDoc / Sphinx: documentation-only (noop_doc executor); calls return the documented signature and snippet, not a live invocation.

The Tool Search pattern (two meta-tools regardless of how many packs you install) and SHA-256 artifact verification apply to every pack regardless of source kind.

• Fetch docs from URLs, GitHub repos, or local paths and index them locally with SQLite FTS5.
• No vector database, no embedding model downloads, and no external API calls on the docs-RAG core path.
• Stores normalized sections in SQLite and writes extracted markdown/json files under .docmancer/extracted/ for inspection.
• Supports GitBook, Mintlify, generic web crawl, GitHub markdown, local directories, and plain text/markdown files.
• Returns compact context packs with estimated token savings and source attribution.
• Installs MCP packs from a local registry directory (DOCMANCER_REGISTRY_DIR) today; the hosted Supabase registry client is implemented in pipeline + registry-api but not yet wired into the CLI install path.

---

API MCP packs

The runtime is wired end-to-end and ships in the PyPI build, but the hosted pack registry that would make install-pack <pkg>@<ver> work out of the box is not yet wired into the CLI. Today you point install-pack at a local registry directory built by the pipeline/ repo:

# Build a pack locally (requires the pipeline checkout) and point the CLI at it.
export DOCMANCER_REGISTRY_DIR=/path/to/local-registry
docmancer install-pack [e-mail]
# Active tool surface: 8 (mode=curated; full=8)
# Required env vars: STRIPE_API_KEY
# Wire-pinned header: Stripe-Version: 2026-02-25.clover
# Destructive endpoints: 2 (gated)
# To enable: docmancer install-pack [e-mail] --allow-destructive

export STRIPE_API_KEY=sk_test_...
docmancer mcp doctor    # verify SHA-256 of every artifact + credential resolution
docmancer mcp list      # inspect installed packs and per-pack state

Inside your agent (Claude Code, Cursor, Claude Desktop, etc.), tools/list always returns just two meta-tools (docmancer_search_tools, docmancer_call_tool), regardless of how many packs you install. The agent searches across every installed pack, then dispatches the resolved tool. Per-call destructive gating, idempotency-key auto-generation, fingerprint-cache reuse on retry, four-source credential resolution, and SHA-256 artifact verification all run inside the dispatcher. The full end-to-end flow lives in the workspace under docs/api-mcp/stripe-walkthrough.md (it is a workspace doc, not a file inside the docmancer/ repo); the runnable demos under docs/api-mcp/demo/ exercise it against a mocked Stripe wire.

---

Commands

Command	What it does
docmancer setup	Create config/database and install detected agent skills
docmancer setup --all	Non-interactively install all supported agent integrations
docmancer add <url-or-path>	Fetch or read documentation and index normalized sections
docmancer update	Re-fetch and re-index all existing docs sources
docmancer query <text>	Return a compact markdown context pack
docmancer query <text> --format json	Return the same context pack as JSON
docmancer query <text> --expand	Include adjacent sections around matches
docmancer query <text> --expand page	Include the full matching page, subject to the token budget
docmancer list	List indexed docsets or sources
docmancer inspect	Show SQLite index stats and extract locations
docmancer remove <source>	Remove a source or docset root
docmancer remove --all	Remove everything indexed (keeps the config)
docmancer doctor	Check config, SQLite FTS5, index stats, and agent skill installs
docmancer fetch <url> --output <dir>	Download docs to markdown files without indexing
docmancer init	Create a project-local docmancer.yaml
docmancer install <agent>	Manual skill installation for a single agent
docmancer install-pack <pkg>@<ver>	Install a version-pinned API MCP pack (--expanded, --allow-destructive, --allow-execute)
docmancer uninstall <pkg>[@<ver>]	Remove an installed pack
docmancer mcp serve	Stdio MCP server bridging installed packs to your agent
docmancer mcp list	Show installed packs, mode, destructive gate state
docmancer mcp doctor	Verify pack SHA-256s, credentials, and agent registrations
docmancer mcp enable|disable <pkg>	Toggle per-pack visibility without reinstalling

---

Retrieval Shape

By default, query uses a 2400 token budget and returns markdown with a summary like:

Context pack: ~900 tokens vs ~4800 raw docs tokens (81.2% less docs overhead, 5.33x agentic runway)

The savings are estimates, but the direction is explicit: compress docs overhead so the remaining token budget goes into useful agent work.

---

Workflow

# 1. Add the docs your agent should see
docmancer add https://docs.pytest.org
docmancer add ./docs

# 2. Install a skill into your agent
docmancer install claude-code

# 3. Query from the CLI or from the agent
docmancer query "How do I use fixtures?"

All agents you install share the same local SQLite index.

---

Keeping Docs Up To Date

Run docmancer update to refresh all locally-added sources. Docmancer re-fetches each URL or re-reads each local path and updates the index in place.

---

Project-Local Config

Global config is stored under ~/.docmancer/ by default. To use a project-local index:

docmancer init
docmancer add ./docs

The generated docmancer.yaml points to .docmancer/docmancer.db and .docmancer/extracted inside the project. If no project config exists, docmancer falls back to the global config.

index:
  db_path: .docmancer/docmancer.db
  extracted_dir: .docmancer/extracted/

---

Supported Agents

setup detects common agent installations. Manual installation remains available:

docmancer install claude-code
docmancer install claude-desktop
docmancer install codex
docmancer install cursor
docmancer install cline
docmancer install gemini
docmancer install github-copilot
docmancer install opencode

Claude Desktop receives a zip package that can be uploaded through Claude Desktop's Skills UI.

---

Optional Extras

Extra	Enables
docmancer[browser]	Playwright-backed fetcher for JS-heavy sites
docmancer[crawl4ai]	Alternative fetcher for hard-to-scrape sites

---

Quickstart | Wiki | PyPI | Changelog