AI Agents with MCP Tools

AllSource's default tenant MCP registry exposes 55 tools to Claude Desktop and other MCP clients. Read-only mode exposes 45 tools; fleet and administrative controls raise the full registry to 73.

What is MCP?

MCP is Anthropic's open protocol for AI tool use. It lets AI assistants call external tools and APIs in a standardized way. all.source implements an MCP server that exposes our entire API as tools.

Available Tools

Event Management

• ingest_event - Create a new event
• query_events - Search events with filters
• reconstruct_state - Reconstruct entity state
• event_timeline - Inspect an entity's ordered history

State and history

• reconstruct_state - Rebuild entity state from ordered events
• get_snapshot - Read a stored entity snapshot
• analyze_changes - Explain changes over a time range

Schema Management

• register_schema - Add a JSON schema
• validate_schema - Validate a schema definition
• list_schemas - Browse the schema registry

Analytics

• cohort_analysis - Analyze cohorts over event history
• correlation_analysis - Find related event patterns
• forecast_events - Forecast event volume from history

Claude Desktop setup and current release status

The stable event-store connector release is 0.22.0. It works with a local or self-hosted Core, but it does not send the Authorization header required by the hosted gateway. The fix is merged for 0.23.0 but is not released yet. Do not use the 0.22.0 image against hosted production: every tool call will return 401 even when CORE_API_KEY is configured.

Until 0.23.0 is published, use the authenticated HTTPS API for hosted data or build the connector from current main after reviewing the unreleased changes. The current MCP setup guide tracks the release boundary and local configuration.

Example Conversations

You: "Show me all user signups from last week"

Claude: Uses query_events to search for user.signed_up events

"Found 847 signup events from the past 7 days. The peak was Tuesday with 156 signups. Here's the distribution by day..."

You: "Summarize event counts by type"

Claude: Uses quick_stats

"Here are the current event counts by type for this tenant."

You: "What was the order state when the payment failed?"

Claude: Uses reconstruct_state and event_timeline

"At the moment of payment failure (2026-02-05T10:30:00Z), the order state was: status='pending', items=[...], total=$127.50. The payment attempt was the 3rd event in this order's history."

Why This Matters

Traditional APIs require developers to write integration code. With MCP, AI agents can:

• Explore your data autonomously
• Reconstruct state and inspect timelines
• Run ad-hoc analytics
• Debug issues with natural language

Your event store becomes a conversational interface.

Get Started

The default connector runs as a local stdio subprocess. Release 0.22.0 supports local and self-hosted Core; hosted gateway support requires the unreleased 0.23.0 authorization fix. Use the authenticated HTTPS API for hosted production until that image is published.

Read the current MCP setup guide or start a hosted trial.