MCP Tools Overview

Wake Intelligence provides 19 MCP tools for temporal intelligence across 5 layers. All tools are accessible through Claude Desktop or any MCP-compatible client.

Current version: v3.6.0 — semantic search quality fixes, admin_reindex_all, score-threshold Vectorize filtering

Tool Categories

Core Context Management

Essential tools for saving, loading, and searching contexts.

Tool	Purpose	Layer
save_context	Save conversation context with AI enhancement	All 4
load_context	Retrieve contexts for a project	Layer 2
search_context	Semantic search via Vectorize (≥0.6 score) with tokenized keyword fallback	Layer 2

Layer 1: Causality (Past - WHY)

Tools for understanding decision history and causal relationships.

Tool	Purpose
reconstruct_reasoning	Explain WHY a context was created
build_causal_chain	Trace decision history backwards
get_causality_stats	Analytics on causal relationships
get_cross_project_dependents	BFS graph of all downstream dependents across projects

Layer 2: Memory (Present - HOW)

Tools for managing memory tiers and access patterns.

Tool	Purpose
get_memory_stats	View memory tier distribution
recalculate_memory_tiers	Update tier classifications
prune_expired_contexts	Clean up old contexts

Layer 3: Propagation (Future - WHAT)

Tools for prediction and pre-fetching optimization.

Tool	Purpose
update_predictions	Refresh prediction scores
get_high_value_contexts	Retrieve likely-needed contexts
get_propagation_stats	Analytics on predictions

Layer 4: Meta-Learning (Adaptive - HOW WELL)

Tools for per-project weight tuning and semantic search reindexing.

Tool	Purpose
get_learning_stats	View adaptive prediction weights per project
reindex_project	Backfill Vectorize embeddings for a single project
admin_reindex_all	Backfill Vectorize embeddings across all projects in one call

Layer 5: Observability + Rune Integration (v3.5.0)

Tools for causal graph visualization, health diagnostics, and Rune Protocol integration.

Tool	Purpose
get_causal_graph	Full project causal network as nodes + edges (D3/Mermaid ready)
get_memory_health	Consolidated 5-layer health report in one call
ingest_rune_manifest	Import `rune.schema.json` — saves each `?` annotation as a Wake causal context

Personality Modes

All retrieval tools (load_context, search_context) accept an optional personality_mode parameter that shapes how results are ranked and presented:

Mode	Ranking	Best For
`historian`	Newest first (default)	Day-to-day use, catching up after a break
`prophet`	By prediction score	Planning what to work on next
`archaeologist`	Most dormant first	Surfacing forgotten threads, long-gap resumption
`minimalist`	Newest first, no framing	Programmatic use, clean output
`auditor`	Grouped by author type	Governance review — human vs. AI-agent vs. AI-compositor

typescript

// Surface what you've been ignoring
load_context({ project: "my-project", personality_mode: "archaeologist" })

// What should I work on next?
load_context({ project: "my-project", personality_mode: "prophet" })

// Who saved what?
load_context({ project: "my-project", personality_mode: "auditor" })

Quick Examples

Save a Context

Claude, save this context:
"Completed database migration 0004 for Layer 3. All prediction columns added successfully."

Project: wake-intelligence
Action type: implementation

Load Recent Contexts

Claude, load contexts for project "wake-intelligence"

Reconstruct Reasoning

Claude, why did we create context [context-id]?

Check Memory Stats

Claude, show me memory statistics for "wake-intelligence"

Update Predictions

Claude, update predictions for project "wake-intelligence"

Get High-Value Contexts

Claude, what contexts am I most likely to need next for "wake-intelligence"?

Tool Design Principles

All Wake Intelligence tools follow these principles:

1. Observable Inputs

Every parameter is based on observable, measurable data:

Project names (strings)
Context IDs (UUIDs)
Time windows (hours, days)
Score thresholds (0.0-1.0)

2. Semantic Outputs

Results include human-readable explanations:

Memory tier names (ACTIVE, RECENT, ARCHIVED, EXPIRED)
Action types (decision, implementation, refactor)
Prediction reasons (recently_accessed, causal_chain_root)

3. Composable Operations

Tools can be chained together:

1. get_high_value_contexts → [list of IDs]
2. load_context → [context details]
3. reconstruct_reasoning → [decision history]

4. Bounded Results

All queries have sensible limits to prevent overwhelming responses:

Default limit: 10 results
Maximum limit: 100 results
Pagination support (coming soon)

Integration Examples

Claude Desktop Workflow

markdown

## Daily Standup

Claude, do these in sequence:

1. Load contexts for project "daily-standup" from the last 24 hours
2. Show me memory stats to see what's ACTIVE
3. Update predictions so we can prefetch tomorrow's likely contexts
4. Get high-value contexts (score > 0.7) to prepare for tomorrow

Automated Context Management

markdown

## Weekly Cleanup

Claude:

1. Show memory stats for all projects
2. Prune expired contexts (older than 30 days)
3. Recalculate memory tiers for all contexts
4. Update predictions for active projects

Causal Analysis

markdown

## Decision Audit Trail

Claude:

1. Search for contexts with tag "architecture-decision"
2. For each result, build the causal chain
3. Reconstruct reasoning to understand WHY decisions were made
4. Show causality stats to identify patterns

Advanced Usage

Prediction-Based Workflow

Use Layer 3 predictions to optimize your AI workflows:

markdown

# Morning Routine

Claude:

1. Get high-value contexts (score > 0.8) for today's projects
2. Load those contexts proactively
3. Check propagation stats to see prediction accuracy
4. If accuracy is low, update predictions with fresh data

Memory Tier Management

Leverage automatic tier classification:

markdown

# Optimize Storage

Claude:

1. Get memory stats for all projects
2. Identify projects with > 50 EXPIRED contexts
3. Prune expired contexts for those projects
4. Recalculate tiers to refresh classifications

Trace decision history:

markdown

# Architecture Review

Claude:

1. Search for "database-migration" contexts
2. Build causal chain from latest migration
3. Reconstruct reasoning for each step
4. Identify which decisions led to current architecture

Tool Response Format

All tools return structured JSON responses following MCP protocol:

json

{
  "content": [
    {
      "type": "text",
      "text": "Human-readable summary"
    },
    {
      "type": "text",
      "text": "Detailed results in markdown"
    }
  ]
}

Error Handling

Wake Intelligence provides clear error messages:

json

{
  "error": {
    "code": "CONTEXT_NOT_FOUND",
    "message": "Context with ID 'abc-123' does not exist",
    "details": {
      "contextId": "abc-123",
      "searchedIn": "wake-intelligence"
    }
  }
}

Performance Considerations

Tool Execution Time

Tool Category	Typical Response Time
Simple queries (load, search)	< 100ms
Layer 1 analysis (causal chains)	100-300ms
Layer 2 stats (memory analytics)	200-500ms
Layer 3 predictions (batch updates)	500ms-2s
Layer 4 meta-learning (weight update)	200-600ms
Semantic search (Vectorize)	50-150ms
Semantic search (LIKE fallback)	< 50ms

Rate Limits

WAF rate limit enforced at the Cloudflare edge (before the worker):

Burst: 30 requests per 10 seconds
On breach: blocked for 10 minutes (per IP)

For personal/team MCP use this limit is well above normal traffic. Heavy batch operations like admin_reindex_all run as a single request internally and are not affected.

Next Steps

Explore individual tool documentation for detailed parameters — start with save_context
Learn about the 5-Layer Brain Architecture — Past / Present / Future / Adaptive / Personality
Run admin_reindex_all if semantic search isn't returning results yet
Check out API Reference for programmatic access

Ready to dive deep? 🐦

Explore Individual Tools →

MCP Tools Overview ​

Tool Categories ​

Core Context Management ​

Layer 1: Causality (Past - WHY) ​

Layer 2: Memory (Present - HOW) ​

Layer 3: Propagation (Future - WHAT) ​

Layer 4: Meta-Learning (Adaptive - HOW WELL) ​

Layer 5: Observability + Rune Integration (v3.5.0) ​

Personality Modes ​

Quick Examples ​

Save a Context ​

Load Recent Contexts ​

Reconstruct Reasoning ​

Check Memory Stats ​

Update Predictions ​

Get High-Value Contexts ​

Tool Design Principles ​

1. Observable Inputs ​

2. Semantic Outputs ​

3. Composable Operations ​

4. Bounded Results ​

Integration Examples ​

Claude Desktop Workflow ​

Automated Context Management ​

Causal Analysis ​

Advanced Usage ​

Prediction-Based Workflow ​

Memory Tier Management ​

Causal Chain Navigation ​

Tool Response Format ​

Error Handling ​

Performance Considerations ​

Tool Execution Time ​

Rate Limits ​

Next Steps ​

MCP Tools Overview

Tool Categories

Core Context Management

Layer 1: Causality (Past - WHY)

Layer 2: Memory (Present - HOW)

Layer 3: Propagation (Future - WHAT)

Layer 4: Meta-Learning (Adaptive - HOW WELL)

Layer 5: Observability + Rune Integration (v3.5.0)

Personality Modes

Quick Examples

Save a Context

Load Recent Contexts

Reconstruct Reasoning

Check Memory Stats

Update Predictions

Get High-Value Contexts

Tool Design Principles

1. Observable Inputs

2. Semantic Outputs

3. Composable Operations

4. Bounded Results

Integration Examples

Claude Desktop Workflow

Automated Context Management

Causal Analysis

Advanced Usage

Prediction-Based Workflow

Memory Tier Management

Causal Chain Navigation

Tool Response Format

Error Handling

Performance Considerations

Tool Execution Time

Rate Limits

Next Steps