MCP Integration¶
Model Context Protocol (MCP) is an industry-standard protocol for connecting AI assistants to external services. Archivus supports MCP as both a client and server, enabling seamless integration with your existing tools.
What Is MCP?¶
Team+
Model Context Protocol:
- Open standard for AI tool integration
- Bidirectional communication
- Secure credential management
- Health monitoring and failover
Think of MCP as:
- USB for AI assistants
- Plug-and-play tool integration
- No custom code required
- Standard interface for all tools
MCP as Client¶
Connect Archie to external MCP servers:
Supported Integrations¶
GitHub
- Create and manage issues
- Search repositories
- Manage pull requests
- Access code and commits
Slack
- Send notifications
- Search message history
- Post updates to channels
- Manage workspace
Databases
- Query internal databases
- Run reports
- Update records
- Aggregate data
Custom APIs
- Any MCP-compatible service
- Internal tools and systems
- Third-party platforms
- Legacy systems
Example Integration: GitHub¶
User: "Create a GitHub issue for the bugs in the QA report"
Archie:
1. Searches for "QA report" in documents
2. Extracts bug descriptions using AI
3. Creates GitHub issues via mcp_github_create_issue
4. Links issues back to source document
Result: Created 3 GitHub issues:
- #142: Login form validation error
- #143: Dashboard loading timeout
- #144: Export CSV formatting issue
All issues tagged with source: QA-Report-2026-02.pdf
MCP as Server¶
Team+
Expose Archivus capabilities to external AI systems:
Available Tools¶
Document Operations
archivus_search_documents- Search document libraryarchivus_get_document- Retrieve document contentarchivus_upload_document- Upload new documentarchivus_analyze_document- AI analysis
Knowledge Graph Enterprise
archivus_query_graph- Search entities and relationshipsarchivus_validate_claim- Verify factsarchivus_get_entity- Retrieve entity details
Workspace Management
archivus_list_workspaces- Get available workspacesarchivus_create_workspace- Set up new workspacearchivus_get_summary- AI-generated overview
Analytics
archivus_get_stats- Usage and performance metricsarchivus_generate_report- Create custom reports
Use Cases¶
Claude Desktop Integration
{
"mcpServers": {
"archivus": {
"command": "npx",
"args": ["-y", "@archivus/mcp-server"],
"env": {
"ARCHIVUS_API_KEY": "your-api-key",
"ARCHIVUS_TENANT_ID": "your-tenant-id"
}
}
}
}
Custom AI Application
from mcp import ClientSession
from mcp.client.stdio import stdio_client
# Connect to Archivus MCP server
async with stdio_client("npx", ["-y", "@archivus/mcp-server"]) as client:
async with ClientSession(client[0], client[1]) as session:
# Search documents
result = await session.call_tool(
"archivus_search_documents",
{"query": "contract expiring Q1 2026"}
)
# Analyze results
for doc in result.content:
analysis = await session.call_tool(
"archivus_analyze_document",
{"document_id": doc.id}
)
MCP Server Configuration¶
Team+
Adding MCP Servers¶
Via Web Interface:
- Navigate to Settings → Integrations → MCP
- Click "Add MCP Server"
- Configure connection:
- Server name
- Command/endpoint
- Environment variables
- Credentials
Via API:
POST /api/v1/mcp/servers
{
"name": "github",
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": {
"GITHUB_TOKEN": "ghp_xxxxx"
}
}
Security Features¶
Credential Encryption
- AES-256-GCM encryption
- Credentials never logged
- Encrypted at rest in database
- Automatic key rotation support
Tenant Isolation
- MCP servers scoped per tenant
- No cross-tenant access
- Separate credential storage
- Independent health monitoring
Health Monitoring
- 5-minute health check intervals
- Automatic failover on errors
- Status dashboard
- Alert notifications
DAG Workflow Integration¶
Team+
Use MCP tools in workflows:
- id: validate_po
type: mcp_tool
config:
server: erp_system
tool: validate_purchase_order
params:
po_number: "{{extracted_po_number}}"
vendor: "{{extracted_vendor}}"
- id: create_issue
type: mcp_tool
config:
server: github
tool: create_issue
params:
title: "{{bug_title}}"
body: "{{bug_description}}"
labels: ["bug", "from-archivus"]
Common Integration Patterns¶
Document Upload Trigger¶
name: external_document_processing
trigger: mcp_server_webhook
nodes:
- id: upload_to_archivus
type: mcp_tool
config:
server: archivus
tool: archivus_upload_document
- id: analyze
type: ai_analyze
- id: notify_slack
type: mcp_tool
config:
server: slack
tool: post_message
params:
channel: "#documents"
text: "New document processed: {{document_name}}"
Cross-System Sync¶
name: crm_document_sync
trigger: document_upload
nodes:
- id: extract_contact
type: ai_extract
config:
fields: [name, email, company]
- id: update_crm
type: mcp_tool
config:
server: salesforce
tool: upsert_contact
params:
email: "{{extracted_email}}"
document_url: "{{archivus_document_url}}"
Research + GitHub¶
name: research_to_issues
trigger: manual
nodes:
- id: research
type: research_task
config:
topic: "{{research_topic}}"
mode: power
- id: extract_action_items
type: ai_extract
config:
fields: [action_items, owners, priorities]
- id: create_issues
type: mcp_tool
config:
server: github
tool: create_issue
params:
title: "{{action_item}}"
assignee: "{{owner}}"
labels: ["from-research"]
MCP Server Status¶
Monitor MCP server health:
Dashboard View:
MCP Servers Status
┌──────────────────────────────────────────────────────┐
│ GitHub Integration │
│ Status: ● Healthy │
│ Last Check: 2 minutes ago │
│ Requests (24h): 147 │
│ Success Rate: 99.3% │
└──────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────┐
│ Slack Integration │
│ Status: ● Healthy │
│ Last Check: 1 minute ago │
│ Requests (24h): 56 │
│ Success Rate: 100% │
└──────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────┐
│ ERP System │
│ Status: ⚠️ Degraded │
│ Last Check: Failed 30 seconds ago │
│ Requests (24h): 234 │
│ Success Rate: 87.2% │
│ Action: [Retry] [Disable] [View Logs] │
└──────────────────────────────────────────────────────┘
Audit Logging¶
Team+
Complete audit trail of MCP operations:
Logged Events:
- Tool invocations
- Parameters and responses
- Success/failure status
- Execution duration
- User attribution
Example Log:
{
"timestamp": "2026-02-07T15:30:45Z",
"user_id": "user-uuid",
"server": "github",
"tool": "create_issue",
"params": {
"title": "Bug in login form",
"repo": "company/product"
},
"response": {
"issue_number": 142,
"url": "https://github.com/company/product/issues/142"
},
"duration_ms": 847,
"status": "success"
}
Cost Tracking¶
MCP tool usage tracked for analytics:
- Calls per server/tool
- Success/failure rates
- Average response times
- Credit consumption (if applicable)
Use Cases by Industry¶
Software Development
- GitHub issue creation from documents
- Automated testing trigger
- Code review automation
- Deploy pipeline integration
Customer Support
- Slack notifications for documents
- CRM contact updates
- Ticket system integration
- Knowledge base sync
Finance
- ERP purchase order validation
- Banking API integration
- Payment system triggers
- Accounting software sync
Healthcare
- EHR patient matching
- Lab system integration
- Insurance verification
- Scheduling system updates
Getting Started¶
- Navigate to Settings → Integrations → MCP
- Add MCP server with credentials
- Test connection
- Use in Archie conversations or workflows
- Monitor status and logs