Skip to content

deepadata/deepadata-edm-mcp-server

Repository files navigation

deepadata-edm-mcp-server

npm version License: UNLICENSED MCP Compatible

The significance layer for AI memory — as an MCP server.

Eight tools for extracting, governing, and querying emotionally significant moments from text. Built on the Emotional Data Model (EDM) open standard.

Quick Install

npx deepadata-edm-mcp-server

Add to Claude Desktop config:

{
  "mcpServers": {
    "deepadata": {
      "command": "npx",
      "args": ["-y", "deepadata-edm-mcp-server"],
      "env": {
        "DEEPADATA_API_KEY": "dda_live_...",
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

What It Does

EDM extracts what mattered from text — not just what happened. Given a journal entry, therapy session, or conversation, it returns structured significance fields: arc_type, emotional_weight, identity_thread, anchor, wound, bridge, narrative.

These fields power significance-weighted retrieval — finding memories by what mattered, not keyword frequency.

The Eight Tools

Extraction & Governance

Tool Description
extract_from_content Extract EDM artifact from text via /v1/extract
seal_artifact Cryptographically seal artifact via /api/v1/issue
validate_edm Validate artifact against EDM v0.8.0 schema
edm_project Project artifact fields for agent context

Significance Routing

Tool Description
deepadata_activate Translate NL query to EDM field filters via /v1/activate

Significance Wiki (new in v0.2.0)

Tool Description
edm_wiki_generate Generate two-file wiki from source text
edm_wiki_search Search local wiki by arc_type, emotional_weight, identity_thread
edm_wiki_lint Health check — arc distribution, recurring threads, temporal span

Significance Wiki

The wiki tools implement the EDM Significance Wiki Format — a two-file pattern for building personal knowledge bases weighted by what mattered.

Given source text, edm_wiki_generate produces:

wiki_article.md — factual: what happened, who was involved, when and where.

significance_article.md — structural: why it mattered, encoded as EDM fields.

Example significance_article.md output

# Significance — journal-entry — 2026-04-08

## Arc Classification
- **arc_type:** threshold
- **emotional_weight:** 0.80
- **valence:** negative
- **recurrence_pattern:** chronic

## Core Significance
- **anchor:** breakup aftermath
- **wound:** inability to be present
- **fuel:** river solace
- **bridge:** walking home the long way
- **identity_thread:** emotional avoidance

## Narrative
After months of absence, she stood in
the coffee shop line looking lighter.
The last kitchen conversation replayed —
her voice steady, saying she needed
presence I couldn't give.

None of those fields appeared in the source text. The structure finds them in the meaning.

Pair with qmd for significance search

edm_wiki_generate produces significance articles
qmd indexes them
Your agent queries by what mattered
# Generate wiki from diary directory
npx deepadata-edm-mcp-server wiki generate ./diary/

# Search by arc type
npx deepadata-edm-mcp-server wiki search --arc-type grief

# Health check your wiki
npx deepadata-edm-mcp-server wiki lint

Significance Routing

deepadata_activate translates natural language queries into EDM significance field filters:

"when was I most afraid of losing her"
        ↓
{ arc_type: "grief",
  emotional_weight: ≥0.7,
  tether_type: "person" }

On significance-typed queries, EDM field routing hits 83.3% recall vs 33.3% for raw vector similarity. The +50pp gap is on queries where "what mattered" is the only path to the right answer.

Environment Variables

Variable Required Purpose
DEEPADATA_API_KEY Yes DeepaData API access
ANTHROPIC_API_KEY Yes (wiki tools) Wiki article generation
DEEPADATA_API_URL No Override API URL
KIMI_API_KEY No Project tool context

Get your API key at deepadata.com

What's New in v0.3.0

  • EDM v0.8.0 support — +gratitude +authenticity arc_types
  • partner: prefix meta.profile per ADR-0017
  • deepadata-edm-sdk ^0.8.4 peer dependency

What's New in v0.2.0

  • edm_wiki_generate — source → significance wiki
  • edm_wiki_search — query local wiki by EDM fields
  • edm_wiki_lint — trajectory pattern detection
  • deepadata_activate — significance routing
  • BYOK: bring your own Anthropic API key

Open Standard

The EDM schema and wiki format are open:

The implementation (extraction intelligence, arc signatures, activation routing) is commercial.

Related

About

The significance layer for AI memory — as an MCP server. Eight tools: extract, seal, activate, wiki.

Resources

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages