MCP SERVER · OFFICIAL PROJECTGive an agent a Markdown knowledge graph it can restructure safely — extract a section into its own note, rename a document, inline a reference — with every link updated as it goes.
LABSCO SUMMARY
What it takes off you is the bookkeeping that normally stops people restructuring: renaming a note, splitting a section out of one, or folding a reference back in all rewrite the links across the graph in the same operation. Reads are shaped for a model rather than a person — retrieval expands outward from what it found, under explicit caps on how many documents and tokens come back.
IWE's own MCP server over a directory of Markdown files. The files stay yours — plain .md, versioned with git — while the server reads them as a graph: documents addressed by key, connected by block references, with every structural operation repairing the links it affects.
npx on your PATH, and a notes directory that has had iwe init run in it. The server reads the directory it is started in, so set the MCP entry's working directory to that folder — otherwise it has no graph to work on. Documents are addressed by an explicit key rather than by title, the attach call resolves its targets from configured attach actions, and mutating queries must carry an expectation, so a query that matched something unintended fails instead of applying.
Follow the published steps
USE CASES
List the sections, extract the one that has outgrown its parent into a standalone document, and a block reference stays behind in its place so nothing loses its context.
A key change rewrites every reference and inline link that pointed at the old one, and the stats call shows what is dangling afterwards.
Fuzzy text is one filter among several: search by what a document references, by what references it, or by a structural set expression with a depth limit.
Create the document at a stable key, then attach it into the targets your configured actions resolve — a daily note, say — with retries left idempotent.
The query carries an expectation of what it should match, so a selector that drifted fails instead of rewriting documents you did not mean.
Flatten the document and everything it references into one Markdown file, ready to paste or publish.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
iwe_attachAttach a document as a block reference in one or more target documents determined by configured attach actions. Each target key is derived from the action's key_template (e.g. daily/{{today}}). The to field accepts a list of action names; the source is attached under each resolved target. Targets that already contain the source are silently skipped. Use list mode to discover available attach actions.
iwe_createCreate a document at an explicit key from the complete markdown in content — frontmatter block first (when there is one), then the title heading and the body. The content is written verbatim; the server adds nothing and moves nothing. Derive the key from stable metadata (entity name, session date), not the title wording. Pass if_exists: "skip" to make retries idempotent. Stats warnings (dangling links, orphans, similar pages) may ride the result; resolve them before ending the session.
iwe_deleteDelete a document from the knowledge graph. All block references and inline links to this document in other documents are cleaned up. Stats warnings (dangling links, orphans, similar pages) may ride the result; resolve them before ending the session.
iwe_extractExtract a section from a document into a new standalone document. The original section is replaced with a block reference. Use list mode to discover sections first
iwe_findSearch and discover documents in the knowledge graph. Supports fuzzy text query (query), root filter (roots), direct-reference filters (refs_to, refs_from), and the structural set selector (in / in_any / not_in / max_depth) for transitive sub-document AND/OR/NOT queries with configurable depth.
iwe_inlineReplace a block reference with the actual content of the referenced document. Use list mode to discover block references first
iwe_normalizeNormalize all document formatting across the knowledge graph. Re-parses and re-writes all documents to ensure consistent formatting
iwe_queryRun an IWE query/block-selection operation document. find and count read; update mutates frontmatter and blocks (operators $replace, $replaceText, $insertBefore, $insertAfter, $append, $delete); delete removes documents. Membership uses the $content filter operator; reads project $content narrowing, $blocks, and $matches. Always strict: every mutating application must carry an expect guard (document-level expect plus one per block operator). Use find with $blocks/$matches to locate targets and learn counts before mutating. Update/delete results may carry stats warnings (dangling links, orphans, similar pages); resolve them before ending the session.
iwe_renameRename a document key. All block references and inline links across the entire graph are updated to point to the new key
iwe_retrieveRetrieve documents from the knowledge graph. Reads the given keys (or search seeds) and expands the graph around them via expand (includes / includedBy / references / referencedBy). With search/fuzzy, seeds are found by relevance within the candidate set (keys + selector); limit caps the seeds before expansion and max_documents caps the documents returned after expansion.
iwe_squashExpand all block references into a single flat markdown document. Useful for export or generating a complete view of a document tree
iwe_statsGet comprehensive statistics about the knowledge graph including document counts, reference patterns, broken links, and most connected documents
iwe_treeView the hierarchical tree structure of the knowledge graph showing how documents are connected via block references. Supports the structural set selector (in / in_any / not_in / max_depth) — when provided, the tree roots are restricted to (or selected from) that set.
iwe_updateUpdate the full markdown content of an existing document. Stats warnings (dangling links, orphans, similar pages) may ride the result; resolve them before ending the session.
BEFORE YOU INSTALL
14 tools when we connected. The CLI commands in the documentation (new, extract, inline, rename and more) are a separate surface; new and export have no iwe_* tool counterpart.
All 3 documented clients run the server on your machine.
CONTEXT COST
Rated by the size of the 14 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Install the IWE CLI — with npm below, or brew install iwe-org/iwe/iwe (macOS/Linux), or cargo install iwe iwes iwec:
npm install -g @iwe-org/iweThen, inside your notes directory, initialise it as an IWE workspace:
iwe initnpx -y @iwe-org/mcp{
"mcpServers": {
"iwe": {
"command": "npx",
"args": [
"-y",
"@iwe-org/mcp"
],
"cwd": "~/notes"
}
}
}Set cwd to the absolute path of your notes directory — the one you ran iwe init in. The server serves the directory it runs in, and clients do not expand ~.
MCP servers
Epistates
Semiotronika
flowing-abyss