TikaAgenticAI
AGENTIC AI AND MCP FRAMEWORK
One AI Engine. Three Access Channels. Unlimited Flexibility.
Tika's agentic AI doesn't just answer questions — it autonomously calls live CRM services to retrieve real-time data. The same tool layer powers the in-app chat, MCP protocol, and insight analysis engine — giving you flexibility to connect any AI client to your CRM intelligence.
TRUSTED BY LEADING LIFE SCIENCES ORGANIZATIONS
ACCESS CHANNELS
Three Ways to Talk to Your CRM
The same AI-powered tool layer is exposed through three distinct channels — each designed for different users and use cases, all querying live CRM data.
Natural language questions from MSLs and reps. The LLM reasons over the question, autonomously calls CRM tools, and streams a grounded answer — territory-scoped and role-restricted.
For Field TeamsOpen standard (Model Context Protocol) that lets any MCP-compatible AI agent — Claude Desktop, Cursor, custom agents — discover and call Tika's CRM tools without custom integration.
For AI Agents & Dev ToolsSpecialized RAG-powered mode that analyzes your entire field insight corpus — surfacing themes, sentiment trends, strategic patterns, and generating downloadable reports.
For Leadership & StrategyHOW IT WORKS
Function Calling on Live CRM Data
Not a chatbot on stale snapshots. The LLM autonomously calls the same Java service methods your REST API uses — always current, always territory-scoped.
"Find psychiatry speakers in New York with CAR-T expertise"
Understands intent, selects the right tool(s), constructs parameters
Calls live CRM service methods with territory-scoped user context
Real-time SSE response grounded in actual CRM data
MCP TOOL CATALOG
11 Live CRM Tools Available Today
Each tool is registered with both the in-app chat assistant (Spring AI @Tool) and the MCP server (@McpTool) — a single implementation, dual exposure. All tools are read-only and idempotent.
Shared Tools — All Applications
Search for HCPs and Accounts by name, specialty, or NPI. Returns matching results with territory membership flags.
Full HCP profile — name, specialty, credentials, NPI, tier, KOL stage, affiliations, territory info, interaction history, and contact details.
Search speakers by topic expertise, specialty, location, tier (Global/Regional/Emerging), and status. Returns expertise tags, rating, event history.
Full speaker profile — credentials, specialty, tier, status, topic expertise, compensation rates, W-9 compliance, event count, rating, and availability.
Paginated interaction list — meetings, calls, emails with HCPs. Filter by HCP name, status, type, date range, and MIR-related flag.
Full interaction record — attendees, products discussed, key topics, notes, MIR data, and associated documents.
Pharma-Specific Tools — Added Per-Request by App Code
Top-scored HCPs in a territory with composite score (0–100) and recommended next best action per HCP. Sorted by priority.
Exact TikaScore for a single HCP — composite score (0–100), per-signal breakdown, and signal weights.
HCPs with declining NRx growth or market share loss in a territory, sorted by worst decline first.
HCPs below 50% call plan attainment and those with engagement decay — not visited recently or at risk of disengagement.
A dedicated RAG-powered tool that analyzes the full corpus of field insights, surfaces themes across territories/TAs, summarizes sentiment, and generates downloadable reports (Word/Excel). Available through the chat assistant's insight analysis mode.
IN ACTION
Real Questions. Real Tools Called.
The LLM autonomously selects and chains tools based on natural language. Here's what that looks like in practice.
Find psychiatry KOLs in the Northeast with active clinical trials
Who are my top 10 HCPs by TikaScore and what should I do next with each?
Find speakers with CAR-T expertise in California who are available
Which HCPs in my territory have declining prescriptions this quarter?
Show me Dr. Smith's full profile, recent interactions, and speaker status
Am I behind on my call plan? Which HCPs haven't I visited recently?
ARCHITECTURE & FLEXIBILITY
Built to Extend
The framework is designed so that adding a new AI capability takes minutes — not weeks. Here's what makes it flexible.
Every CRM tool is written once and registered on both the chat assistant (@Tool) and MCP server (@McpTool). Same service layer, same territory scoping, dual access.
Every tool call resolves user context server-side from the authenticated session — never trusting LLM-provided IDs. RBAC and territory restrictions apply identically to AI and REST.
Shared tools load for all apps. Pharma-specific tools (TikaScore, NBA, call plan) are added per-request based on the user's app code — MSL users never see pharma capabilities.
Supports OpenAI GPT-4o and Anthropic Claude, configurable per deployment. The tool layer is LLM-agnostic — switch providers without changing a single tool.
Answers stream token-by-token via Server-Sent Events. Multi-turn conversation context maintained within sessions for follow-up questions.
Any MCP-compatible AI agent (Claude Desktop, Cursor IDE, custom agents) can discover and call Tika's tools without custom integration. API-key authenticated.
EXTENSIBILITY
Adding a New Tool Is This Simple
The framework is designed to make adding new AI capabilities trivial. Here's the pattern — every tool follows it.
Same service layer used by REST controllers. Territory-scoped, tested, production-ready.
Add description and parameter annotations. The LLM reads these to decide when to call the tool. MCP clients auto-discover it.
The tool is instantly available to the in-app chat, MCP clients, and any future AI channel. No wiring, no config changes.
QUESTIONS
Common questions
Is this a chatbot on exported CRM data?
No. The model calls the same live service methods the REST API uses. The answer is only as current as the CRM record.
What is MCP doing here?
MCP is the open protocol that lets another AI client discover and call Tika’s CRM tools. The in-app chat uses the same tools.
Can an MSL see pharma-only tools?
No. Shared tools load for every app. TikaScore, next best action, and call-plan tools load from the user’s app code.
Does the model pick which HCP IDs it is allowed to see?
No. Territory and role are resolved server-side from the authenticated session.