Overview
The MCP Server exposes Athena’s core capabilities as standardized Model Context Protocol tools, making them consumable by any MCP-compatible client including Antigravity, Claude Desktop, Cursor, and other agentic IDEs.Quick Start
stdio Transport (IDE Integration)
For local IDE integration using standard input/output:SSE Transport (Remote Access)
For remote or multi-client access using Server-Sent Events:IDE Configuration
Add Athena to your IDE’s MCP settings (e.g.,.agent/mcp_config.json):
Available Tools
The MCP server exposes 8 tools, each gated by the Permissioning Layer:| Tool | Permission | Sensitivity | Description |
|---|---|---|---|
smart_search | read | internal | Hybrid RAG search with RRF fusion |
quicksave | write | internal | Save timestamped checkpoint to session log |
health_check | read | public | Audit Vector API and Database subsystems |
recall_session | read | internal | Retrieve recent session log content |
governance_status | read | internal | Check Triple-Lock compliance state |
list_memory_paths | read | public | List active memory directories |
set_secret_mode | admin | — | Toggle demo/external mode |
permission_status | read | — | Show permission state and tool manifest |
smart_search
Hybrid RAG search combining Canonical Memory, Tags, Vectors, GraphRAG, and filenames:quicksave
Save a checkpoint to the current session log:The quicksave tool automatically checks Triple-Lock compliance and includes governance status in the response.
health_check
Run a health audit of core services:Resources
The server exposes 2 resources accessible via URI:| URI | Description |
|---|---|
athena://session/current | Full content of active session log |
athena://memory/canonical | Canonical Memory (CANONICAL.md) |
Permissioning Layer
All tools are gated by the Permission Engine (see Security for details).Capability Tokens
4 escalating permission levels:| Level | Access | Example Tools |
|---|---|---|
read | Query/read data | smart_search, recall_session |
write | Modify session logs | quicksave |
admin | Modify config, toggle modes | set_secret_mode |
dangerous | Delete data (future, unused) | — |
write (can access read + write tools)
Sensitivity Labels
3 data classification tiers:| Label | Description | Examples |
|---|---|---|
public | Safe for demos, external sharing | Health check, memory paths |
internal | Normal operational data | Session logs, search results |
secret | Credentials, finances, PII | API keys, trading data |
Secret Mode
Toggle withset_secret_mode(True) when:
- Screen sharing during demos
- External pair-programming
- Client presentations
- ✅
health_checkandlist_memory_pathsremain accessible (PUBLIC) - 🔒 All INTERNAL/SECRET tools are blocked
- 📝 Remaining data sources auto-redact sensitive content (
[REDACTED])
Architecture
The MCP server acts as a thin transport layer, exposing Athena SDK capabilities through the standardized MCP protocol while enforcing security policies via the Permission Gate.Implementation Reference
Seesrc/athena/mcp_server.py:58 for the smart_search tool implementation and src/athena/core/permissions.py for the permission gating logic.
Dependencies
The MCP server is included when you install Athena from source with
pip install -e .Next Steps
Security Model
Learn about permissioning, secret mode, and data residency
Vector RAG
Understand the hybrid search architecture