MCP Protocol Integration

Bidirectional Tool Connectivity for Limitless Extensibility

Overview

Model Context Protocol (MCP) is Anthropic’s open standard for connecting AI systems to external tools and data sources. Archivus implements MCP bidirectionally, functioning as both a client (consuming external tools) and a server (exposing capabilities to other systems).

Archivus as MCP Client

Connect external MCP servers to extend Archie’s capabilities.

Supported Integrations

Service	Capabilities	Use Case
GitHub	Issues, PRs, code search	“Create an issue for this bug”
Slack	Messages, channels	“Post summary to #updates”
PostgreSQL	SQL queries	“Query customer data”
Filesystem	File operations	“Read config files”
Custom	Any MCP server	Your integrations

Configuration

POST /api/v1/mcp/servers
Content-Type: application/json

{
  "name": "GitHub Integration",
  "server_type": "http",
  "endpoint_url": "https://mcp.github.example.com",
  "tool_prefix": "github",
  "auth_type": "bearer",
  "credentials": {
    "token": "ghp_xxxxxxxxxxxx"
  },
  "enabled": true
}

Tool Namespacing

External tools are prefixed to avoid collisions:

Archie's 26 built-in tools:
├── search_documents
├── ask_documents
├── compare_documents
└── ...

+ MCP GitHub tools:
├── mcp_github_create_issue
├── mcp_github_list_repos
└── mcp_github_search_code

+ MCP Slack tools:
├── mcp_slack_post_message
├── mcp_slack_search
└── ...

Using MCP Tools in Archie

Once connected, use natural language:

“Create a GitHub issue titled ‘Documentation outdated’ in the archivus repo with the summary from this document”

Archie automatically selects and invokes the appropriate MCP tool.

Archivus as MCP Server

Expose Archivus capabilities to other AI systems.

Available Tools

When connected as an MCP server, Archivus exposes:

Tool	Description
`archivus_search`	Semantic document search
`archivus_ask`	Q&A with document context
`archivus_summarize`	Document summarization
`archivus_upload`	Document ingestion
`archivus_list`	Document listing
`archivus_analyze`	Document analysis

MCP Server Endpoint

GET /api/v1/mcp/server/info

Response:
{
  "name": "archivus",
  "version": "1.0.0",
  "protocol_version": "2024-11-05",
  "capabilities": {
    "tools": true,
    "resources": true,
    "prompts": false
  },
  "tools": [
    {
      "name": "archivus_search",
      "description": "Search documents by semantic meaning",
      "inputSchema": {
        "type": "object",
        "properties": {
          "query": {"type": "string"},
          "limit": {"type": "integer", "default": 10}
        },
        "required": ["query"]
      }
    }
  ]
}

Connecting External AI Systems

External AI systems can connect to Archivus MCP server:

# Example: Claude Desktop connecting to Archivus
{
  "mcpServers": {
    "archivus": {
      "url": "https://api.archivus.app/mcp",
      "auth": {
        "type": "bearer",
        "token": "YOUR_API_KEY"
      }
    }
  }
}

MCP in DAG Workflows

Use external MCP tools within DAG orchestration.

MCP Node Type

{
  "node_type": "mcp_tool",
  "config": {
    "server_id": "github_integration",
    "tool_name": "create_issue",
    "parameters": {
      "repo": "",
      "title": "",
      "body": ""
    }
  }
}

Example: Document to GitHub Issue

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Trigger   │────▶│  Summarize  │────▶│   Create    │
│  (Upload)   │     │  Document   │     │GitHub Issue │
│             │     │  (AI Node)  │     │ (MCP Node)  │
└─────────────┘     └─────────────┘     └─────────────┘

Security Model

Credential Encryption

All MCP credentials are encrypted:

AES-256-GCM encryption at rest
Credentials never logged
Per-tenant key isolation
Automatic key rotation (Enterprise)

Tenant Isolation

MCP connections are tenant-scoped:

Each tenant manages their own MCP servers
No cross-tenant tool access
Audit logging for all MCP operations

Permission Model

{
  "mcp_permissions": {
    "can_add_servers": ["admin"],
    "can_use_tools": ["admin", "editor"],
    "can_view_servers": ["admin", "editor", "viewer"]
  }
}

Transport Types

HTTP (Recommended)

Standard HTTP-based transport for production:

{
  "server_type": "http",
  "endpoint_url": "https://mcp.example.com",
  "timeout_ms": 30000
}

SSE (Server-Sent Events)

For real-time streaming responses:

{
  "server_type": "sse",
  "endpoint_url": "https://mcp.example.com/sse",
  "keepalive_ms": 30000
}

Stdio (Development Only)

For local development with subprocess-based servers:

{
  "server_type": "stdio",
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"]
}

Health Monitoring

Connection Health

GET /api/v1/mcp/servers/{id}/health

Response:
{
  "server_id": "github_integration",
  "status": "healthy",
  "last_check": "2026-01-18T10:30:00Z",
  "latency_ms": 145,
  "tools_available": 12,
  "recent_errors": 0
}

Usage Statistics

GET /api/v1/mcp/servers/{id}/stats

Response:
{
  "server_id": "github_integration",
  "period": "last_24h",
  "total_calls": 156,
  "successful_calls": 154,
  "failed_calls": 2,
  "avg_latency_ms": 230,
  "by_tool": [
    {"tool": "create_issue", "calls": 45},
    {"tool": "search_code", "calls": 89}
  ]
}

API Reference

MCP Server Management

Endpoint	Method	Description
`/mcp/servers`	GET	List MCP servers
`/mcp/servers`	POST	Add MCP server
`/mcp/servers/{id}`	GET	Get server details
`/mcp/servers/{id}`	PUT	Update server
`/mcp/servers/{id}`	DELETE	Remove server
`/mcp/servers/{id}/test`	POST	Test connection
`/mcp/servers/{id}/toggle`	PUT	Enable/disable
`/mcp/servers/{id}/tools`	GET	List server tools

Archivus MCP Server

Endpoint	Method	Description
`/mcp/server/info`	GET	Server capabilities
`/mcp/server/tools`	GET	Available tools
`/mcp/server/invoke`	POST	Invoke tool

Tier Availability

Feature	Team	Enterprise
MCP Client (connect to servers)	5 servers	Unlimited
MCP Server (expose to others)	Read-only	Full access
MCP in DAG Workflows	Full	Full
Custom MCP Servers	-	Available

Best Practices

Security

Use HTTPS - Always use secure endpoints
Rotate Credentials - Regular credential rotation
Minimal Permissions - Grant only required access
Monitor Usage - Review MCP audit logs

Performance

Use HTTP Transport - Avoid stdio in production
Set Timeouts - Appropriate timeout configuration
Enable Health Checks - Monitor connection status
Cache Where Possible - Reduce redundant calls

Organization

Descriptive Names - Clear server naming
Consistent Prefixes - Logical tool namespacing
Document Integrations - Maintain integration docs

Ready to extend Archivus? Configure MCP Integrations →