Understanding MCP Server

The Model Context Protocol (MCP) is an open standard that enables ADM to connect seamlessly with external tools, services, and data sources. Think of MCP as a universal translator — it standardizes how ADM communicates with diverse systems across your data ecosystem, from cloud storage to catalogs, documentation platforms, and orchestration tools.

How MCP Works

MCP acts as a bridge between ADM and external services. When you request information from an external system, the query travels through the MCP framework as follows:

User Query → ADM Agent → MCP Server → External Service

                   ↓                        ↓

               Response ← MCP Protocol ← Service Data

ADM sends standardized requests via the MCP protocol. Each configured MCP server translates these into the API language of the target system, retrieves the response, and formats it into a structure ADM can interpret and present.

Key Benefits of MCP Integration

MCP vs. Traditional Integrations

With MCP, ADM becomes a single pane of glass across your data stack — combining operational insights from ADOC with business and governance context from external tools.

Configuring MCP Servers

Accessing MCP Configuration

Go to Settings > Integrations > MCP Servers to view and manage all MCP connections. The page displays each server’s name, URL, connection status, and available tools.

Health Status Indicators

Server URL Requirements

All MCP endpoints must follow standardized suffixes:

This ensures ADM can establish proper communication channels with each MCP server.

Connecting Through MCP Hubs

You can connect directly to individual servers or use MCP hubs like Composio, which aggregate multiple integrations into a single access point.

Benefits of MCP Hubs:

• Single authentication point for all tools
• Centralized credential management and monitoring
• Consistent API patterns
• Pre-built connectors to popular platforms (Google Drive, Slack, Atlassian, etc.)

Example: Using Composio as an MCP Hub

1. Log in to Composio.
2. Create a project and configure integrations (e.g., Google Drive, Slack).
3. Generate an endpoint URL ending with /mcp or /sse.
4. Add this endpoint in ADM — one connection grants access to all integrated tools.

Adding a New MCP Server

1. Go to Settings > Integrations > MCP Servers.
2. Click + Add Server.
3. Fill in the required fields:

Example URL

https://apollo.composio.dev/v3/mcp/5add3917.../mcp?user_id=pg-test-6f1cd49f...

4. Click Add Server.
5. Once connected, you’ll see:
   1. Connected status badge
   2. Available Tools link showing capabilities (e.g., “Available Tools (30)”)
   3. Last Checked timestamp
   4. Action buttons: Test Connection, Edit, Delete

Managing Servers

• View Details: Click a server card to see connection info and available tools.
• Edit: Update server name, URL, or description via the edit icon.
• Test Connection: Use the refresh icon to verify connectivity; ADM updates the status badge and timestamp.
• Remove: Click delete > confirm. (Conversations using this server will lose access to its tools.)

Important Notes on MCP Behavior

MCP responses may vary due to:

• The LLM model used
• Query phrasing
• Server capabilities and available data
• Conversation context

Tip: Treat the examples in this documentation as guides — your actual ADM responses will adapt to your environment, connected systems, and ongoing conversation context.

Google Drive Integration for Data Governance

Google Drive integration enables ADM to read documents, spreadsheets, and files from your organization’s Drive — empowering data teams to bridge business documentation and operational policies.

Common use cases include:

• Translating BRDs into data quality policies
• Validating existing policies against documented standards
• Enriching data quality investigations with business impact context

Setup via Composio

Step 1: Configure in Composio

1. Log into composio.dev.

2. Create or open a project > MCP Config > Create MCP Config.

3. Click Add Toolkit > Google Drive.

4. Set up authentication:

   • Choose OAuth.
   • Select required permissions (file read/search).
   • Complete Google authorization.

5. Click Create MCP > Create Instance.

6. Copy the generated MCP URL (ends in /mcp or /sse).

Step 2: Add to ADM

1. In ADM: Settings > Integrations > MCP Servers > + Add Server

2. Enter:

   • Server Name: “GoogleDocs”
   • Server URL: Paste the Composio endpoint
   • Description: “Access to Google Drive documents”

3. Click Add Server and verify CONNECTED status.

Example Use Case 1: Creating Policies from Business Requirements

You: “Read the Q4_2025_Data_Requirements.pdf from my Drive and identify what data quality requirements we need.”

ADM:

I found 12 data quality requirements across customer and financial data. Recommending policies such as:

• Email format validation for customers.email_address
• Conditional phone number checks for premium customers
• Completeness validation on address fields
• Transaction amount range and reconciliation checks

ADM extracts policy logic, severity, and affected tables directly from your documents, enabling pre-populated policy creation in Workflow Mode.

Workflow Integration Benefits:

• Auto-fills rule logic and thresholds
• Adds BRD references and requirement IDs
• Aligns severity with business impact
• Maintains traceability between policy and documentation

Example Use Case 2: Validating Policies Against Documentation

You: “Compare our current data quality policies with the standards in Data_Governance_Framework_2025.pdf.”

ADM:

47 policies analyzed.

• 32 compliant with framework
• 8 missing required rules (e.g., date range validation)
• 2 conflicting configurations (e.g., uniqueness constraints)

Recommendations:

• Create missing policies for uncovered requirements
• Adjust thresholds to match governance standards
• Align documentation and operational rules

Example Use Case 3: Enriching Business Context

You: “The customer_ltv column in customer_360_view is showing anomalies. What’s the business impact?”

ADM:

According to Customer_Analytics_Playbook.pdf:

• LTV drives marketing budgets ($2.3M quarterly)
• Influences CS resource allocation and recommendation models
• Current anomalies could misdirect campaigns worth ~$180K

Recommended Actions:

1. Pause affected automations
2. Correct source data and reprocess
3. Implement policy for order amount anomalies

This integration allows ADM to connect technical findings with business impact, turning investigations into actionable intelligence.

Capabilities & Privacy

What ADM Can Do

• Search and read files (Docs, Sheets, PDFs, text)
• Navigate folders and shared drives
• Retrieve metadata (owner, timestamps)

What ADM Cannot Do

• Create, delete, or edit files
• Modify permissions or upload content

Privacy

• ADM accesses files only when queried
• No file content is stored
• Access is logged with timestamps and user identity
• Permissions can be revoked anytime via Google or ADM

Atlassian Integration (Confluence & JIRA)

The Atlassian integration connects ADM to Confluence (knowledge/wiki) and JIRA (issue tracking) via MCP. This enables:

• Pulling governance standards, runbooks, and business definitions from Confluence into conversations and workflows.
• Correlating ADOC quality incidents with JIRA issues, and creating or updating tickets directly from context.

Setup via Composio (Recommended)

Step 1: Configure in Composio

1. Sign in to composio.dev > MCP Config > Create MCP Config.

2. Add Toolkit > select Atlassian (Confluence + JIRA) or add them individually.

3. Choose OAuth > complete Atlassian authorization.

4. Select scopes:

   • Confluence: read pages, search content, access spaces
   • JIRA: read issues, search projects, view issue details

5. Create MCP > Create Instance > authorize your Atlassian site.

6. Copy the generated MCP URL (ends with /mcp or /sse).

Step 2: Add to ADM

1. ADM > Settings > Integrations > MCP Servers > + Add Server
2. Name (e.g., “Atlassian”), paste Server URL, add a description.
3. Confirm status shows CONNECTED > open Available Tools to verify.

Use Cases

Enrich lineage with business context (Confluence)

You: “Show complete lineage for customer_360_view, including business context from Confluence.”

ADM: Merges technical lineage (ADOC) with business lineage and consumers (Confluence), highlighting purpose, SLAs, owners, and downstream impact.

Why it matters: Aligns technical impact with business-critical dependencies (e.g., campaign spend, CS assignments, executive dashboards).

Validate standards vs. policies (Confluence → ADOC)

You: “Review our Financial Data Quality Standards page and tell me what policies we should have.”

ADM: Extracts standards (completeness, reconciliation, ranges, timestamp logic, referential integrity), proposes policy types, rules, severities, and target tables.

Workflow boost: Pre-populates policy forms with rule logic and links to Confluence for traceability.

Correlate incidents with tracked work (JIRA)

You: “Any open JIRA issues related to orders data quality problems?” ADM: Matches ADOC issues with JIRA tickets (status, assignee, comments), identifies gaps (no ticket yet), and offers to create new issues with context.

Outcome: Prevents duplicates, speeds remediation, and keeps audit trails aligned.

Capabilities & Permissions

Note: Actions like ticket creation depend on the MCP server exposing those tools in a read/write mode. Many deployments start read-only.

Data Catalog Integration (Alation)

Alation provides business context (owners, glossary, classifications, rules, lineage). Through MCP, ADM+ADOC can combine operational quality metrics with catalog governance to drive prioritized action.

Setup

• Via Composio: Add the Alation toolkit if available, configure API token, create an instance, copy /mcp URL, add to ADM.
• Direct MCP: If your Alation offers a native MCP endpoint, add it directly (must end with /mcp or /sse).

Confirm CONNECTED and verify Available Tools.

Use Cases

Policy recommendations from catalog rules

You: “Review Alation’s business rules for customer_orders and propose ADOC policies.” ADM: Returns policy list (names, rules, severity, rationale), mapped to Alation rules and tables. Value: Bridges documentation and enforcement; keeps policies faithful to business intent.

Prioritize by criticality

You: “Show ADOC quality status for Alation ‘Tier 1 Critical’ assets.” ADM: Correlates classification with quality scores, flags business impact (owners, consumers, SLAs), and prioritizes fixes (e.g., SOX-tagged tables).

Close expectation gaps

You: “Find mismatches between Alation expectations and ADOC policies.” ADM: Highlights incomplete or missing policy coverage (e.g., freshness SLA not enforced, referential integrity partial), with actionable remediation plan.

Best Practices

• Link policies to Alation pages for audit traceability.
• Use Alation severity/classification to set policy severity and alerting.
• Review quarterly: ensure catalog rules are actually enforced in ADOC.

Validating Documentation vs. Reality

Documentation drifts. ADM can compare documented facts (Confluence/Alation) with actual state (ADOC) and produce precise gaps.

Checks include:

• Documented-but-missing assets
• Outdated descriptions, owners, schedules
• Missing business context (purpose, consumers, SLAs)
• Quality expectations documented but not enforced

Outputs:

• Gap tables (undocumented assets; incomplete metadata)
• Risk assessment (usage, downstream consumers, compliance tags)
• One-click actions (create JIRA tickets, draft page updates, schedule recurring validations)

Enterprise Custom MCP Servers

Why Custom MCP

Many enterprises have internal systems (homegrown catalogs, observability, compliance tools, legacy platforms). A custom MCP server exposes these systems to ADM via a standard, secure protocol — unlocking unified access without bespoke connectors.

Technical Requirements

Protocol

• Implement Model Context Protocol (v1.0+), JSON-RPC 2.0 over HTTP/SSE
• Endpoints must end with /mcp or /sse
• Return well-formed MCP messages and capability schemas

Network

• HTTPS (TLS 1.2+, ideally 1.3)
• Stable, reachable endpoint (support health checks)

Auth

• API key or bearer token (stateless per request)
• Clear error messages on auth failures
• Minimum-privilege service accounts

Capabilities

• Declare tools with input/output schemas, version them, and document rate limits and errors.

Configuration in ADM

1. Prepare: URL, credentials, capability list, SLOs, contacts.
2. ADM > Settings > Integrations > MCP Servers > Add MCP Server
3. Enter Name, URL, Description, Auth (API key header or bearer token).
4. Test Connection > verify capabilities discovered.
5. Enable > server available in conversations/workflows.

Example Capability Set (Internal DataHub)

• search_datasets(query, filters?) → list datasets with metadata
• get_dataset_metadata(dataset_id) → full metadata
• get_lineage(dataset_id, direction, depth) → graph or adjacency
• get_business_rules(dataset_id) → rules for policy mapping
• get_usage_stats(dataset_id, time_range) → adoption signals

Usage:

• Find assets by owner/domain, fetch rules, generate ADOC policies, and link documentation for audit.

Security & Operations

Authentication & Authorization

• 32+ char API keys/tokens; rotate quarterly
• Scope tokens minimally; never hardcode secrets
• Enforce resource-level authorization server-side

Network & Abuse Protection

• HTTPS only; optional VPN/IP allowlists
• WAF + rate limiting (e.g., 100–500 req/min)
• Monitor anomalies (geo/time anomalies, spikes)

Data Handling

• Return minimal, masked data; avoid PII by default
• Log access with user + timestamp; retain logs ~90 days
• No sensitive caching in responses

Compliance

• Document accessible data domains
• Maintain audit trails and change logs
• Review access monthly; patch regularly

Reliability

• Health endpoints; connection pooling; timeouts (30s), retries (3, backoff)
• Versioned APIs; backward-compatible changes
• Non-prod test before prod rollout

Monitoring & Maintenance

Deployment Models

• Hub-only (Composio): Centralized auth/monitoring for many external tools.
• Direct custom servers: Fine-grained control for internal systems.
• Hybrid (Recommended): Hub for standard SaaS; direct MCP for proprietary systems.

Practical Tips

• Start read-only. Add write actions after controls are proven.
• Document capabilities with examples and SLA expectations.
• Instrument metrics from day one; build a health dashboard.
• Automate tests for schema, auth, and rate-limit behavior.