Shelfmark README
The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Shelfmark listing page.
shelfmark
Give AI agents the right context, not your entire filesystem.
Shelfmark turns years of scattered documents into a governed context map for AI agents. An agent can discover what exists, understand what kind of material it is, and select the documents relevant to a task — before spending context opening files.
No document migration. No duplicated content store. No need for a perfect folder structure.
Local by design · metadata only · governed discovery · built for MCP
Your best knowledge is probably sitting in your folders
Reports, presentations, models, research, proposals and working documents accumulate over years. Some are carefully organised. Others sit inside crowded project folders, old archives, download directories, or collections that made sense only at the time.
The value is still there. The problem is that agents cannot use what they cannot discover — and giving an agent unrestricted filesystem access does not solve that. It transfers the work of finding, interpreting and filtering thousands of files into the context window.
Shelfmark gives the agent a map first.
Context is the scarce resource
An agent does not need every document. It needs to know what exists, what is likely to matter, where it came from, and whether it should be used at all.
Shelfmark is a discovery layer between the agent and your files. The agent searches the catalogue, narrows the field, and requests only the material the task actually needs — so an existing document estate becomes working context without turning the filesystem into one enormous prompt.
A catalogue of pointers, not another document store
Shelfmark builds a local SQLite catalogue of references and derived metadata: paths, filenames, formats, sizes, document types, authors, dates, classifications, selected Office properties, presentation titles, and optional content hashes.
It does not copy your documents into the catalogue. It does not index document body text. It does not create a second repository to govern, synchronise and maintain. Your files stay where they are; the catalogue points at them and describes what can be established from their metadata.
Finding a document does not expose its contents.
Turn messy folders into usable agent context
Shelfmark does not require a designed information architecture. Point it at accumulated project files, forgotten archives, or folders where documents have simply been dropped over the years.
It builds an inventory from signals that already exist — filenames, extensions, paths, authorship, dates, Office properties, and your own classification rules — creating a virtual structure across the material. Agents can then search by document type, client, project, purpose, author, year or location without anything being moved or renamed.
A messy folder can become navigable even when it never becomes tidy.
Shelfmark also stays honest about the limits of metadata. A file called
final7.pdf with no useful properties cannot reveal its meaning without
someone reading it. Shelfmark reports what is known rather than inventing
certainty — the same reason it drops placeholder slide titles instead of
listing twelve headings a deck does not have.
Built for selective context
- Discover before opening. Search thousands of references before deciding which few files deserve attention.
- Preserve the context window. Concise catalogue results instead of whole documents that may not be relevant.
- Build on previous work. Reports, models, proposals, research and methods stay discoverable across new tasks and future engagements.
- Keep provenance visible. Path, date, author, document type and surrounding context help an agent judge relevance.
- Reduce blind exploration. Structured search instead of repeatedly walking directories and inspecting files one at a time.
- Separate discovery from access. Shelfmark helps identify material; opening the original remains a separate, controllable decision.
Your roots are the trust boundary
Shelfmark walks only the roots you configure.
Symlinks are not followed. A link inside a root reads as an ordinary
file and would otherwise walk straight out of the tree you configured — and
hash opens files. Skipped links are reported, never silently dropped. To
index another tree, add it as an extra root: the boundary widens by saying
so in config, not by planting a link.
The catalogue is refused inside any root, because a database that indexes
itself grows on every refresh. Both checks compare resolved paths, so ..
and a symlinked root cannot slip past them.
Governance belongs in the retrieval layer
Not every useful document should be treated the same way. Shelfmark separates two questions that usually get confused:
- Who owns or may reuse this? →
rights:OWN/REFERENCE/RESTRICTED - May this document leave its current context? →
confidential:0/1
A method may belong to you while the client deliverable containing it stays confidential. Modelling the two separately lets agents discover reusable knowledge without treating everything discoverable as freely shareable.
Files matching your private/secret patterns become RESTRICTED: no tool
returns their path, name, metadata or content, no argument overrides it,
they are never opened for hashing, and the database is opened read-only.
corpus_stats() reports a single corpus-wide count of sealed files and
nothing else about them — not which root, not which folder. That count is
the one thing disclosed, deliberately: silence about it would misrepresent
the size of the corpus.
Governance is applied by the catalogue, not left to the wording of a prompt.
Know whether the map can be trusted
A search result is only useful if the agent knows the catalogue is current.
The MCP server keeps its own index current while it runs, so nothing has to
be scheduled and no agent has to remember. When it cannot — never built,
stale, a failed refresh, a clock it cannot reason from, or an index that no
longer agrees with the filesystem — every tool says so above its answer,
and corpus_stats() compares index against disk in full.
An old snapshot is never presented as complete knowledge.
Designed for knowledge-intensive work
| Consultants and advisors | Find previous analyses, proposals, frameworks and deliverables without exposing unrelated client material. |
| Researchers and analysts | Navigate large collections of reports, datasets and source material through consistent metadata. |
| Product and strategy teams | Reconnect decisions, research, roadmaps and previous thinking across projects and time. |
| Studios and independents | Turn years of accumulated work into reusable context while keeping control over client files and IP. |
| Agent builders | Give local agents a governed discovery layer over MCP. |
How it works
- Point Shelfmark at your existing folders. One or more roots. Files stay where they are.
- Build the local catalogue. It walks the permitted roots, extracts available metadata, applies classification rules, and writes references into SQLite.
- Review ownership and confidentiality.
shelfmark reviewasks a few questions about your own folders and writes the answers to config. - Connect an MCP-compatible agent. It searches, browses and inspects catalogue records through structured tools.
- Retrieve only what matters. The agent identifies the relevant artefacts before any separate content access takes place.
Install
Python ≥ 3.11. macOS, Linux and Windows — the full suite runs on all three in CI, including the Windows-specific behaviours (OneDrive placeholder detection, junction refusal at the root boundary).
Codex plugin
Install the MCP server and its catalogue-management, context-finding, and archive-research skills as one versioned plugin:
Then create the local catalogue once with uvx shelfmark init followed by
uvx shelfmark refresh. The plugin starts the pinned Shelfmark MCP server
over stdio; document paths and metadata stay on the local machine.
To install the three companion skills without the MCP configuration:
Email ingestion is optional, and the extra you want depends on the format
you have. .msg resolves to wheels everywhere; .pst needs
libpff-python, which publishes no wheels and compiles from C source, so
it requires a build toolchain (Visual C++ Build Tools, build-essential,
or the Xcode command line tools):
Quickstart
Three commands, and init finds your documents for you — when the default
root misses, it sweeps for the folders that do hold documents and one
keypress fixes the config:
That's install to connected. The agent's first call then looks like this
(a real corpus_stats() answer over the corpus above):
Optional but worth the five minutes: shelfmark review asks a few
questions about your biggest unclassified subtrees and writes the answers
to config, so rights stop being UNKNOWN; shelfmark stats prints the
census any time; shelfmark config shows every rights rule in the order
it is checked, with the number of files each one currently claims.
Then in a session: corpus_stats() to orient, browse_folder() to
navigate, search_docs() / get_file() to find and inspect.
MCP tools
| Tool | What it answers |
|---|---|
corpus_stats() | What is here overall + an honest freshness line. Call first. |
browse_folder(prefix) | What is inside a folder: counts, sizes, facet mix. |
search_docs(query, …) | Metadata full-text search with facet filters. |
get_file(path) | Full record for one file: rights, authorship, slide titles, identical copies, on-disk status. |
search_emails(query, …) | Full-text over an ingested .pst/.msg email archive (optional). |
Result lists always say when they are cut (showing 100 of 195 …), unknown
filter values are reported as bad filters with suggestions (never as an
empty corpus), and excerpts mark their truncation point.
Configuration
Everything corpus-specific lives in config.toml — the code ships with
neutral defaults only. Resolution order: --config flag →
$SHELFMARK_CONFIG → ~/.config/shelfmark/config.toml. See
config.example.toml for the full annotated reference. Highlights:
| Section | What it controls |
|---|---|
[[roots]] | The trees to index. One unlabelled primary root; extra roots get a label prefix. |
[index] | Where the SQLite catalogue lives. Must be outside every indexed root and outside cloud-synced folders — it is a mutating binary DB, and this is enforced. |
[privacy] | Regexes for secrets and private subtrees → RESTRICTED. Built-ins already cover .env, key/cert files, id_rsa, backup codes, identity documents. |
[authors] | Regexes for your own name/company, for client authors, and for generator tools — drives OWN/REFERENCE classification from OOXML authorship. |
[rights] | Path-prefix rules for the two-axis model: rights (may I reuse it) × confidential (may it leave). |
[facets] | Which top-level folders count as work/personal; where client/project names sit in the path. |
[doc_types] / [context_types] | Extra filename/folder rules, checked before the built-in bilingual (EN/ES) defaults; built-ins can be disabled by name. |
What shareable_only means
shareable_only=True means positively classified: confidential=0 AND rights IN (OWN, REFERENCE). Never-reviewed files are held back — unreviewed
is not the same as cleared.
Getting rights set: shelfmark review
Most files carry no OOXML author — a corpus is mostly PDFs, markdown and
images — so authorship classifies almost nothing and path rules are the
only mechanism that can. They ship empty, which is why a fresh catalogue
is mostly UNKNOWN and shareable_only comes back nearly empty.
That knowledge is yours, not the corpus's, so review asks for it — biggest
win first, with what the catalogue knows on screen:
Answer own, own-private, reference, client, personal, scratch
or skip. On a ~1,900-file corpus, five answers settled 80% of the
unclassified files and eight settled 90%.
It stops whenever you do, only ever asks about subtrees that are still
unclassified (so re-running continues where you left off), and writes
config, never rights values in the data — so any answer is undone by
editing the config and re-running shelfmark rights. Where authorship is
lopsided enough to be evidence it offers a default; where it is thin it
stays quiet rather than guessing.
Editing classification rules
Two things people trip over, learned the hard way:
-
Anchor short regex alternatives on both sides. An unanchored
rfimatches inside "Dockerfile" and the Spanish word "perfil". Before adding or deleting an alternative, list the filenames it actually matches. -
A rule edit does not relabel existing files. The builder is incremental, so after any rule change run:
Do not use
build --rebuildfor this: it re-walks everything (and on cloud-synced trees drags every evicted file back down) when the filenames are already in the DB.
Keeping it current
refresh is incremental by (size, mtime, cloud-residency) — a couple of
seconds over a ~30k-file tree — and it asserts its own correctness on every
run: walk coverage (an OS-denied walk must not read as success), guarded
pruning of deleted files (refused when it looks like a mass deletion or an
unmounted root), and governance invariants (no RESTRICTED row in the search
index, no unsealed secret, no unsearchable non-RESTRICTED row). Results
land in REFRESH_STATUS.json next to the DB, and the MCP server reports
them on every corpus_stats() call.
Any guard that declines to touch the index says so on stderr and exits non-zero — you never get a clean-looking summary hiding a refusal.
When something is wrong and nothing says so
Checks the setup for the failures that stay quiet: a catalogue inside a
cloud-synced folder (a sync client uploads it on every write, and one that
copies the .db, -wal and -shm at different moments restores a torn
database that opens and answers wrongly), a root the process cannot read
— on macOS usually Full Disk Access, which is what an empty walk is hiding
— a database folder that is not writable, a corpus that is mostly
cloud-evicted and what that costs you, and any email format sitting in the
catalogue with no reader installed.
It names the fix rather than the fault, prints the evidence behind each guess so you can overrule it, and exits non-zero on anything fatal so it can gate a setup script.
No flag changes what is checked. The one flag there is changes only who the output is for:
The same verdicts as JSON, with the corpus taken out — counts, states and verdict codes, never a path, a filename, a root label or a config string. Because nothing here phones home, a broken catalogue on your machine is invisible to anyone who could help; this is the one channel a local-only tool gets, and at roughly 1.5 KB it fits in an issue. It carries the failure streak, which is the difference between "it is broken" and "it has been broken since Tuesday, 113 runs".
The whole shape, nothing elided:
The redaction is not a convention anyone has to remember: the report copies no free text, so a check added later cannot leak through it, and a test plants a canary string in the root name, the filename and the config and fails if it comes out the other end.
When a guard stops you
The two size guards cannot tell "the root was unreadable" from "those files really were deleted" — both look like a short walk. So they refuse, name both possibilities, and leave the index untouched:
Check which it was. If the files are genuinely gone:
--force backs the catalogue up to catalog.db.bak-preprune before
deleting anything. If instead a root was merely unmounted or unreadable,
fix that and refresh normally — the rows are still there.
A worked example: somebody renames a folder
This is the failure the guards exist for, start to finish. A 200-file
corpus: 60 documents in Clients/Acme, 140 in Admin. Every line below is
real output; only the paths are shortened.
1 — a clean build.
2 — Clients/Acme is renamed to Clients/ACME Corp. Nothing was
deleted, but every catalogued path under the old name now points at
nothing, and the walk finds 60 files it has never seen:
The refusal is the right call: from here a rename and a mass deletion are
the same event. The status is now degraded, not ok — the run finished,
but the catalogue knowingly lists 60 rows whose files are gone.
3 — run it again and the catalogue is carrying both copies, so the coverage floor takes over from the prune ceiling:
It stays failed on every subsequent run. It does not quieten down.
4 — meanwhile, what an agent gets. No tool answers as if nothing happened; the warning leads, and the streak is in it:
Note the paths: it is still answering, and answering from the old names. That is why the banner is not optional.
5 — ask what is wrong.
6 — the files really did move, so accept it.
7 — confirm, and note that stats ends with a real walk, not a clock:
The one warning left is 100% of the catalogue has UNKNOWN rights, which
is true of any corpus before shelfmark review has run — see
Getting rights set.
You do not schedule this
The MCP server keeps its own index current. It is spawned by your client,
lives for the whole session, and refreshes once on startup and then
whenever refresh.max_age_seconds has passed. Nothing to install, no
timer to configure, no command to remember.
That works because a resident process the client already starts is a
better trigger than a scheduler: it inherits the same file access the
client has, and it is running exactly when you are asking questions. If
you maintain the catalogue some other way, shelfmark-mcp --no-auto-refresh leaves it alone.
Editor hooks are now an optimisation, not the mechanism — they cut the
delay between an agent writing a file and that file being searchable, from
one refresh interval to the end of the turn. With Claude Code, in
~/.claude/settings.json:
mark-dirty drops a marker only when a write landed under an indexed root
(near-free on every other write); hook stop picks it up at the end of the
turn. hook session-start refreshes and then checks the index against the
disk — the same check the MCP server runs — telling you and the agent only
when something is wrong.
Do not wrap these in >/dev/null 2>&1. An earlier revision of this page
taught exactly that, and it is how a catalogue stayed silently wrong for
four days: every refusal and failure the refresh reports goes to stderr,
so the redirect discarded the only delivery of the news. shelfmark hook
is silent when the catalogue is healthy and always exits 0 — there is
nothing left to suppress.
If you also want it current with no client running
Only worth it for non-MCP use, or an instant first query on a very large
tree. On Linux, a systemd --user timer running shelfmark refresh --if-needed is clean.
On macOS, the obstacle is TCC, not launchd. A LaunchAgent is denied
~/Documents, ~/Desktop and ~/Downloads by default; os.walk swallows
the error, so the build walks a handful of files and exits 0 — every
layer reports success while the index never updates. Grant Full Disk Access
to the interpreter that runs the job and a LaunchAgent is fine. The
coverage assertion catches the ungranted case either way: the refresh
fails loudly rather than quietly indexing nothing.
On Windows, Task Scheduler running the same command is the equivalent; no scheduled-run recipe has been field-tested there yet, so if you set one up, the coverage assertion is again what stands between a broken schedule and a silently stale index.
Content hashing and duplicates
The refresh never reads file contents (it must stay fast). To populate hashes for duplicate detection:
Sensitive rows are never opened, and neither are symlinks. On cloud-synced trees, dataless placeholder files are skipped — reading one silently yields the hash of the empty string, which would make distinct files look identical. Re-run after large materialisations.
What makes Shelfmark different
- Desktop search helps a person find text inside files.
- Document management systems require documents to be imported and managed inside a new environment.
- Retrieval systems parse, chunk and copy document content into search indexes or vector stores.
- Filesystem tools give an agent direct access and leave every discovery decision to the agent.
Shelfmark sits at a different layer: a local, governed, metadata-based map of the documents you already have, built specifically for selective agent context.
What Shelfmark does not try to be
It is not a document management system. It does not replace your filesystem. It does not require embeddings. It does not reorganise your folders. It does not claim to understand content it has never read. And it does not assume every discoverable document is safe to share.
It gives agents a better starting point.
These are decisions, not a backlog:
- No embeddings. Metadata FTS + facets first.
- No content extraction. Body text stays out of the index by design.
Letting real misses decide
Those two refusals should be revisited on evidence, not on a competitor's feature list — so shelfmark records the searches that found nothing, locally, and tells you what they mean:
The report answers one question: could metadata search ever have found it? A term appearing in no filename, path, author, title or slide title was unreachable however it was phrased — that is the shape content extraction fixes. A term that is in your corpus but still missed was a phrasing or filter problem, which is a different repair. Without that split, a miss log only proves that people search.
Bad filters, impossible year ranges and searches against a stale index are
not recorded as misses: each is already explained to the caller, and
logging them would bury the real signal. It stays local, is capped, and
never leaves your machine. Turn it off with [misses] enabled = false —
recording your own search terms is a privacy-affecting choice, and whether
that is acceptable genuinely varies by corpus.
FAQ
Does Shelfmark upload my documents? No. It runs locally and builds a local catalogue. Document contents are not copied into it.
What does the catalogue contain? File references and derived metadata: paths, filenames, types, sizes, dates, authors, classifications, selected Office properties, presentation titles, and optional content hashes.
Does it read document contents?
The file catalogue does not index body text. Two deliberate exceptions:
shelfmark hash opens files to compute content hashes for duplicate
detection, and optional email ingestion can index message bodies so
search_emails works. Both are opt-in commands, both skip symlinks, and
both honour your privacy rules — a file or archive matching your
secret/private patterns is never opened by either.
Do I need to reorganise my folders first? No. Better filenames and metadata improve classification, but no formal structure is required.
Can it understand every badly named file? No. A file with a meaningless name and no embedded metadata stays hard to identify without reading it. Shelfmark says what it knows rather than guessing.
Why not just give an agent filesystem access? Filesystem access lets an agent open files. Shelfmark helps it decide which files are worth opening — and which should stay out of its results entirely.
Why not a vector database? A vector database is for semantic retrieval from contents. Shelfmark solves the earlier problem: discovering and governing what exists, before deciding what content should be processed at all. They are compatible; this one comes first.
Which agents can use it?
Any client that supports local MCP servers. Register with
claude mcp add shelfmark -s user -- shelfmark-mcp, or point your client at
the shelfmark-mcp command.
Development
The test corpus — including its OOXML files — is synthesised on every run; no fixture binaries are committed and no real document is ever read.
License
MIT — see LICENSE. Free to use, modify and redistribute.
mcp-name: io.github.Dankaro-projects/shelfmark