Wordcell documentation

Install the CLI, create a vault, save a decision, find it, connect it, and preview a published page.

Set up, query, maintain, and revise repository memory with a coding agent.

Register Wordcell’s local MCP server with Claude Code, Codex, or Cursor, check that the agent can start it, and fix the errors you are likely to see.

Search an Obsidian vault in place, see which links, tags, and attachments Wordcell reads, and serve the vault to a coding agent without changing it.

Export your Supermemory documents and memory entries, import them as Markdown notes, and see what does not transfer.

Save a page, thread, or video as Markdown with local copies of its assets and a record of how it was captured.

Turn a local or remote PDF into a Markdown bundle with the original file, headings, and extracted evidence.

Select notes, inspect the dry-run report, build the site, and choose how to host it.

Search and link across only the vaults you explicitly select and authorize.

Keep one vault current on several machines with a private Git repository and a script that commits and pushes on a schedule.

The source version can reorder up to 25 results with Cloudflare Clef. If the provider fails, you keep the original order.

How maintainers cut a release, and how you can check a downloaded archive's signatures and provenance.

Exact interfaces, SDK imports, optional adapters, the vault format, and prerequisites.

Endpoints, MCP tools, tokens, and limits for hosting a site built from selected notes on wordcell.io.

Publish against an expected revision and recover the exact result after an interrupted request.

Why Wordcell keeps everything in Markdown files, and how its storage, search, graph, and capture fit together.

Why durable agent memory belongs in inspectable Markdown beside the repository.

Markdown knowledge tools and agent memory services compared with Wordcell, using each project's own primary documentation.

Reproduce source-level experiments on excerpt selection and packed context, using the frozen inputs and recorded results.

Run named graph queries, read their proofs and limits, and rebuild the local graph cache.

A historical comparison of exact and hybrid search on a frozen repository corpus, with measurements and uncertainty.