# Edward MCP server

> A local-first research memory and writing workspace that indexes sources, extracts clean text, and provides evidence-backed retrieval for AI agents and humans.

- Repository: https://github.com/nmorton13/edward
- Page: https://hostedmcp.org/servers/nmorton13/edward
- Transport: —
- Stars: 0
- License: MIT

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `EDWARD_ANSWERER_MODE` | no | no |  |
| `EDWARD_ANSWER_BASE_URL` | no | no |  |
| `EDWARD_ANSWER_LOCATION` | no | no |  |
| `EDWARD_ANSWER_MODEL` | no | no |  |
| `EDWARD_SUMMARIZER_MODE` | no | no |  |
| `EDWARD_SUMMARIZER_MODEL` | no | no |  |
| `EDWARD_SUMMARIZER_PROVIDER` | no | no |  |

## FAQ

### How does Edward handle web page and PDF extraction?

Web pages and PDFs are fetched and extracted into clean readable text using the `edward process` command, which can be automated with a scheduler.

### Can I use Edward without an external AI model?

Yes, Edward supports exact lexical search via SQLite FTS5 and can answer questions using a `--no-model` flag to return a structured evidence packet.

### What kind of sources can I ingest into Edward?

You can ingest URLs, local files, text notes, X (Twitter) bookmarks, self-sent Gmail messages, and structured Markdown or JSON research bundles.

### How does Edward ensure citations are accurate?

When an answer model is active, the system verifies every citation against the underlying source text and falls back to a retrieved evidence packet if they cannot be verified.