145 Tools, 17 Skills, 3 Systems, One Session: Inside the MCP Ecosystem That Powers R2-CX

What Is the MCP Ecosystem?

The Model Context Protocol (MCP) is the interface layer between R2-CX's AI reasoning engine and the enterprise systems it operates on. Each MCP server exposes a set of tools that the AI can call with structured parameters and receive structured responses. Think of it as a typed API specifically designed for AI agents: every tool has a schema, every parameter has validation, and every response follows a predictable format.

The JedIN platform runs 14 MCP servers covering SAP, Microsoft, Salesforce, and proprietary systems. Three of these servers are production-validated with 145 tools and 17 skills that have been tested against live systems with real data. This post documents the complete validation results from a testing campaign that executed 35 test scenarios with a 100% pass rate.

The Three Production MCP Servers

MCP C4C (Port 3333) -- 19 Tools

The SAP Cloud for Customer MCP server is unique in the ecosystem because it uses Playwright browser automation (v1.59.1) rather than API calls. SAP C4C does not expose its configuration and fine-tuning capabilities through external APIs, so R2-CX drives a headless Chromium browser to interact with the C4C UI exactly as a human consultant would.

Complete Tool Inventory:

Tool	Category	Description
c4c_login	Authentication	Full browser login with credential entry and session establishment
c4c_navigate	Navigation	Navigate to any C4C work center or view by name or path
c4c_click	Interaction	Click any UI element by selector, text, or accessibility label
c4c_type	Interaction	Type text into input fields with automatic focus handling
c4c_screenshot	Evidence	Capture full-page or element-specific screenshots
c4c_evaluate	Advanced	Execute JavaScript in the browser context (same-origin OData via fetch)
c4c_adaptation_mode	Configuration	Toggle C4C adaptation/personalization mode
c4c_create_field	Configuration	Create custom extension fields on business objects
c4c_fine_tuning	Configuration	Navigate and modify fine-tuning configuration sections
c4c_open_config_section	Configuration	Open specific fine-tuning section by path
c4c_read_table	Data	Extract tabular data from the currently visible UI
c4c_fill_form	Interaction	Fill multiple form fields in a single operation
c4c_select_option	Interaction	Select dropdown values, radio buttons, checkboxes
c4c_wait	Control	Wait for specific UI conditions (element visible, text present)
c4c_scroll	Navigation	Scroll within containers to load lazy-rendered content
c4c_get_page_info	Inspection	Extract current page title, URL, and visible element summary
c4c_close_dialog	Interaction	Close modal dialogs and popups
c4c_generate_absl	Code Gen	Generate ABSL code from 6 templates (validation, calculation, visibility, approval, deduplication, territory)
c4c_analyze	Analysis	Run analysis engines: data_quality, config_completeness, full_audit, report

The c4c_analyze tool runs four distinct analysis types:

• data_quality: Scores data quality across business objects (score: 0-100, benchmark result: 95/100 in testing)
• config_completeness: Evaluates how completely C4C has been configured (benchmark result: 88/100)
• full_audit: Comprehensive system audit covering security, configuration, and data (benchmark result: 70/100)
• report: Generates a 4,000+ character executive markdown report with scores, findings, and recommendations

Performance Benchmarks (C4C) -- Measured from test sessions:

Tool	Avg Response Time	Notes
c4c_login	~9,600ms	Full browser authentication flow
c4c_navigate	1-515ms	Varies by page complexity
c4c_evaluate	4ms	JavaScript execution in browser context
c4c_generate_absl	2ms	Template-based code generation
c4c_screenshot	~180ms	PNG capture and encoding
c4c_open_config_section	109ms-4,900ms	Varies by section depth
c4c_analyze	1,200-3,800ms	Depends on analysis type

MCP CPI (Port 3335) -- 47 Tools

The SAP Cloud Platform Integration MCP server communicates via OData APIs, providing complete access to the CPI management plane. This server covers package management, artifact CRUD, credential operations, monitoring, and deployment.

Tool Categories:

Category	Tool Count	Key Tools
Authentication	1	cpi_login
Package Management	6	cpi_list_packages, cpi_get_package, cpi_create_package, cpi_update_package, cpi_delete_package, cpi_export_package
Artifact Management	8	cpi_list_artifacts, cpi_get_artifact, cpi_create_artifact, cpi_update_artifact, cpi_delete_artifact, cpi_download_artifact, cpi_upload_artifact, cpi_copy_artifact
Deployment	4	cpi_deploy_artifact, cpi_undeploy_artifact, cpi_get_deploy_status, cpi_list_deployed
Monitoring	7	cpi_get_message_logs, cpi_get_message_details, cpi_get_error_info, cpi_get_runtime_status, cpi_get_performance_metrics, cpi_get_trace, cpi_set_trace_level
Credentials	5	cpi_list_credentials, cpi_get_credential, cpi_create_credential, cpi_update_credential, cpi_delete_credential
Certificates	4	cpi_list_certificates, cpi_get_certificate, cpi_upload_certificate, cpi_delete_certificate
Resources	4	cpi_list_resources, cpi_get_resource, cpi_upload_resource, cpi_delete_resource
Configuration	3	cpi_get_config, cpi_update_config, cpi_list_adapters
Utilities	5	cpi_search, cpi_get_system_info, cpi_list_event_types, cpi_get_properties, cpi_validate_artifact

The CPI MCP server achieves near-100% parity with the SAP CPI Web IDE capabilities. Every operation a consultant would perform through the browser-based CPI tooling can be executed programmatically through these 47 tools.

MCP JedIN (Port 3334) -- 79 Tools + 11 Skills

The JedIN MCP server is the largest in the ecosystem, providing complete platform management through REST API calls. It covers everything from flow design to deployment, monitoring, and tenant administration.

Tool Categories:

Category	Tool Count	Key Tools
Authentication	2	jedin_login, jedin_refresh_token
Flow Management	12	jedin_list_flows, jedin_get_flow, jedin_create_flow, jedin_update_flow, jedin_delete_flow, jedin_publish_flow, jedin_deploy_flow, jedin_undeploy_flow, jedin_execute_flow, jedin_get_flow_status, jedin_export_flow, jedin_import_flow
Package Management	6	jedin_list_packages, jedin_get_package, jedin_create_package, jedin_update_package, jedin_delete_package, jedin_export_package
Connection Management	6	jedin_list_connections, jedin_get_connection, jedin_create_connection, jedin_update_connection, jedin_delete_connection, jedin_test_connection
Credential Management	5	jedin_list_credentials, jedin_get_credential, jedin_create_credential, jedin_update_credential, jedin_delete_credential
Monitoring	8	jedin_get_execution_metrics, jedin_list_executions, jedin_get_execution, jedin_get_execution_logs, jedin_get_system_health, jedin_get_performance, jedin_list_alerts, jedin_acknowledge_alert
Tenant & Users	6	jedin_get_tenant_info, jedin_update_tenant, jedin_list_users, jedin_get_user, jedin_create_user, jedin_update_user
Marketplace	5	jedin_list_marketplace, jedin_get_marketplace_item, jedin_install_marketplace, jedin_rate_marketplace, jedin_search_marketplace
Artifacts	8	jedin_list_artifacts, jedin_get_artifact, jedin_create_artifact, jedin_update_artifact, jedin_delete_artifact, jedin_get_artifact_content, jedin_update_artifact_content, jedin_list_artifact_versions
Code Assist	10	Templates, node types, connector patterns, code validation, code generation
Schema	4	jedin_parse_schema, jedin_list_schemas, jedin_validate_schema, jedin_transform_schema
Utilities	7	jedin_search, jedin_get_changelog, jedin_get_audit_log, jedin_get_feature_flags, jedin_get_system_info, jedin_get_documentation, jedin_get_connector_info

11 Skills (Higher-Level Operations):

Skills combine multiple tool calls into guided workflows:

Skill	Tools Used	Description
create-rest-integration	4-6 tools	Create a complete REST-to-REST integration flow
create-file-integration	4-6 tools	Create file-based integration (SFTP, S3)
create-event-integration	3-5 tools	Create event-driven integration with webhooks
create-sap-integration	5-8 tools	Create SAP-specific integration (RFC, IDoc, OData)
get-connector-info	1-2 tools	Retrieve detailed connector documentation
diagnose-flow	3-5 tools	Diagnose a failing flow with log analysis
optimize-flow	2-4 tools	Suggest performance optimizations for a flow
migrate-cpi-flow	4-8 tools	Migrate a CPI iFlow to JedIN format
setup-monitoring	2-3 tools	Configure alerts and dashboards for a flow
bulk-deploy	3-6 tools	Deploy multiple flows with dependency ordering
security-audit	4-6 tools	Audit credentials, connections, and access patterns

Performance Benchmarks (JedIN) -- Measured from test sessions:

Tool	Measured Response Times	Notes
jedin_list_flows	22ms, 24ms, 35ms	Complete flow inventory
jedin_create_flow	69ms	Flow creation with validation
jedin_list_credentials	14ms, 20ms, 28ms	Credential store query
jedin_get_tenant_info	114ms, 137ms, 256ms	Full tenant configuration
jedin_list_users	10ms	User account listing
jedin_get_execution_metrics	43ms, 53ms	Aggregated runtime statistics
jedin_list_node_types	2ms	Node type categories
jedin_get_health	4ms	System health check
jedin_deploy_flow	~120s (polling)	Includes Camel K build + deploy

Auto-Login and Session Recovery

One of the most critical features of the MCP ecosystem is automatic session recovery. Enterprise systems have session timeouts, token expiry, and occasional connectivity interruptions. If the AI consultant's session dies mid-operation, it needs to recover without human intervention.

How Auto-Login Works

Each MCP server implements a token lifecycle manager:

1. Initial authentication: The AI calls the login tool (e.g., jedin_login). The server stores the access token, refresh token, and expiry timestamp.
2. Token refresh: Before each tool call, the server checks if the access token expires within the next 60 seconds. If so, it automatically refreshes using the stored refresh token.
3. Full re-authentication: If the refresh token itself has expired (e.g., after a long idle period), the server performs a full login using stored credentials and retries the original tool call.
4. Transparent retry: The AI receives the tool response as if nothing happened. The re-authentication is invisible to the reasoning layer.

Real-World Session Recovery (Confirmed in Testing)

During the 61-scenario benchmark test, R2-CX encountered session expiry twice:

• Scenario 43: SAP C4C session died after 45 minutes of continuous operation. The MCP server detected the session cookie was invalid ("Token expirado"), performed c4c_login (~9.6s), and retried the failed call. Total recovery time: approximately 10 seconds. Zero manual intervention.
• Scenario 55: JedIN JWT expired during a long analysis phase. The MCP server refreshed the token in milliseconds and retried. The AI did not even notice the interruption.

This auto-recovery capability is essential for the multi-hour consultant sessions that R2-CX supports. A session that requires human intervention to re-authenticate is a session that cannot run autonomously.

Validation Results: 35/35 Tests Pass

The MCP ecosystem was validated through a structured test campaign covering all three production servers:

Test Distribution

Server	Tests	Tool Calls	Pass Rate
MCP C4C	12	45+	12/12 (100%)
MCP CPI	10	30+	10/10 (100%)
MCP JedIN	13	42+	13/13 (100%)
Total	35	117+	35/35 (100%)

Cross-System Session Test

The most demanding test scenario involved all three systems in a single session:

1. Login to SAP C4C (browser), JedIN (REST), and SAP CPI (OData) -- 3 parallel logins
2. Extract C4C configuration data using browser automation
3. Query JedIN for existing integration flows that connect to C4C
4. Query CPI for artifacts that interact with the same C4C tenant
5. Cross-reference findings and generate a unified assessment

This scenario executed 60+ tool calls across 3 MCP servers in a single architect session. All calls succeeded. Total session duration: under 5 minutes for complete cross-system analysis.

The Full Ecosystem: 14 Servers, 307 Tools

Beyond the 3 production servers, the JedIN MCP ecosystem includes 11 additional servers at various stages of readiness:

API-Ready (Awaiting Azure AD Tenant)

Server	Port	Tools	Target System
MCP Power BI	3336	26	Microsoft Power BI
MCP Power Apps	3337	18	Microsoft Power Apps / Dynamics 365

These servers are code-complete and tested against mock endpoints. They await production Azure AD credentials for live validation.

Beta (Functional, Pending Production Testing)

Server	Port	Tools	Target System
MCP Salesforce	3338	18	Salesforce CRM
MCP HubSpot	3339	15	HubSpot CRM
MCP VTEX	3340	14	VTEX Commerce
MCP TOTVS	3341	12	TOTVS Protheus
MCP S/4HANA	3342	16	SAP S/4HANA
MCP ABAP	3343	12	SAP ABAP Systems
MCP SAC	3344	11	SAP Analytics Cloud
MCP Emarsys	3345	11	SAP Emarsys
MCP C4C v2	3346	11	SAP C4C (API-based, next gen)

These beta servers have been developed and tested with mock or sandbox data. They are not yet validated against production systems and are not included in the "145 tools" production count.

External (Official Provider MCPs)

For AWS, Azure, and Dynamics 365, JedIN integrates with official MCP servers provided by the platform vendors (AWS Labs MCP, Azure MCP) rather than building custom implementations.

Total Ecosystem Numbers

Category	Servers	Tools	Skills
Production (validated)	3	145	17
API-Ready	2	44	0
Beta	9	118	0
Total	14	307	17

Note: The 6 additional skills (beyond the 11 JedIN skills) come from the MCP C4C and MCP CPI servers, bringing the total to 17 skills across the 3 production servers.

Architecture: How MCP Servers Connect

Every MCP server follows a shared architecture pattern from the packages/mcp-shared library:

R2-CX AI Engine
    |
    | (MCP Protocol - JSON-RPC over stdio/HTTP)
    |
    +-- MCP Server Base (mcp-shared)
         |
         +-- Tool Registry (schema validation, parameter types)
         +-- Auth Manager (token lifecycle, auto-refresh, re-login)
         +-- Response Formatter (structured output for AI consumption)
         +-- Error Handler (retry logic, graceful degradation)
         |
         +-- Transport Layer
              |
              +-- Playwright (C4C: browser automation)
              +-- HTTP/REST (JedIN: REST API)
              +-- OData (CPI: SAP OData services)
              +-- OAuth2 (Power BI, Power Apps: Azure AD)

The shared base library handles authentication, retry logic, and response formatting. Individual MCP servers only need to implement the transport layer and tool-specific logic. This shared architecture is why the ecosystem scales to 14 servers without proportional engineering effort.

What 145 Validated Tools Means for Enterprise AI

The number itself is less important than what it represents: complete operational coverage across three enterprise systems. R2-CX does not generate generic advice based on documentation. It reads real configurations, queries real APIs, navigates real UIs, and produces findings based on actual system state.

Every tool call documented in this post returned real data from live production and sandbox systems. Every response time is a measured value, not an estimate. Every finding was derived from actual system inspection, not from templates or assumptions.

This is what separates an AI consultant platform from a chatbot with enterprise knowledge: the ability to act, not just advise.