Sales Assistant — Lead-to-Quote MCP Server

Microsoft 365 Copilot · FastMCP · Salesforce CRM · Streamable HTTP · Azure Container Apps

An enterprise-grade Model Context Protocol server that gives fintech sales professionals a complete Lead-to-Quote workflow without ever leaving the flow of work. It powers two surfaces:

1. A Microsoft 365 Copilot Declarative Agent — chat against your Salesforce pipeline with rich, branded HTML widgets, built with the M365 Agents Toolkit (VS Code) and the RemoteMCP plugin.
2. A Copilot Cowork plugin — five co-work skills (research briefs, follow-up automation, pitch decks, meeting scheduling, pipeline review packs) that turn Salesforce signals into finished email, Word, PowerPoint and Excel artifacts. See cowork/.

---

📸 Screenshots

Sales Dashboard	Pipeline Summary

Account 360 (with M365 signals)	Opportunities (Kanban + table)

Leads	Dark mode

Widgets carry the Microsoft 365 Copilot look — a light surface, the signature rainbow accent bar, Segoe UI typography — and adapt automatically to light, dark and high-contrast themes.

---

✨ Features

Capability	Details
Salesforce CRM	Leads, Accounts, Contacts, Opportunities, Quotes, Products, Activities
Authentication	Client Credentials, OAuth Bearer pass-through, Entra SSO (switchable via env)
UI Widgets	Copilot-branded HTML+Skybridge widgets bound to _meta.structuredContent, with light/dark/contrast theming
M365 Integration	Agenda/email helpers for Teams, Outlook via M365 Tooling Servers
Cowork plugin	Five co-work skills that orchestrate Salesforce data with Cowork's built-in email/Word/PowerPoint/Excel
Market Intelligence	Designed to pair with external market-intelligence MCP tools for company, macro and market context
Compliance	MiFID II / GDPR / FCA compliance flags and deal desk approval tracking
Transport	Streamable HTTP (/mcp) or stdio
Cloud	One-command deploy to Azure Container Apps (scripts/deploy-l2q.ps1)

---

🏗 Architecture

Microsoft 365 Copilot
   │
   ├── Declarative Agent ──► RemoteMCP plugin
   │                              │
   └── Cowork plugin ──► remoteMcpServer connector
                                  │
                                  ▼
                  Sales Assistant MCP Server (FastMCP / Streamable HTTP)
                  hosted on Azure Container Apps
                                  │
        ┌─────────────────────────┼─────────────────────────┐
        ▼                         ▼                         ▼
  Salesforce REST API      M365 Tooling Servers       Market-intelligence MCP
  (SOQL)                   Teams · Outlook · Calendar  (external, in agent)
  Leads / Accounts /       (referenced, not proxied)   company · macro · headlines
  Contacts / Opps /
  Quotes / Products

Each tool returns:

• content – Narrative text with prompt suggestions so the agent can continue a natural conversation.
• _meta.structuredContent – JSON data consumed by the HTML+Skybridge widgets at ui://widgets/<name>.html.

Most tools are read-only. Create/update tools (create_record, update_record) are write operations and require explicit confirmation (confirm=true).

---

🚀 Quick Start (local)

1. Prerequisites

• Python 3.11+
• Salesforce Connected App with Client Credentials flow enabled
• (Optional) Entra ID app registration for SSO

2. Install

python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -e .[dev]

3. Configure

Copy-Item env\l2q.example.env env\l2q.env

Edit env/l2q.env:

# Auth mode: client_credentials | oauth_bearer | entra_sso
SF_AUTH_MODE=client_credentials
SF_INSTANCE_URL=https://your-org.my.salesforce.com
SF_CLIENT_ID=3MVG9...
SF_CLIENT_SECRET=...
SF_API_VERSION=v59.0

# Only needed for entra_sso mode
AAD_APP_CLIENT_ID=
AAD_APP_TENANT_ID=

4. Run locally

scripts\run-l2q.ps1
# OR
python -m l2q.cli --transport http --host 0.0.0.0 --port 8080

The server listens at http://localhost:8080/mcp.

5. Test with MCP Inspector

1. Open MCP Inspector → New HTTP Connection → http://localhost:8080/mcp
2. Add header Authorization: Bearer <token> (for oauth_bearer / entra_sso modes)
3. Call get_sales_dashboard to see your morning briefing

---

☁️ Deploy to Azure

The server ships with an idempotent deployment script that builds the image in Azure Container Registry (cloud build) and rolls it out to Azure Container Apps.

az login
.\scripts\deploy-l2q.ps1
# Optional: .\scripts\deploy-l2q.ps1 -Location eastus -ImageTag v1.2.0

It provisions / updates (all idempotent, safe to re-run):

Resource	Name	Notes
Resource Group	rg-l2q-dev
Container Registry	acrl2qdev	Basic SKU
Container Apps Env	cae-l2q-dev	auto-creates Log Analytics
Container App	ca-l2q-dev	scale-to-zero, 0–3 replicas, 0.5 vCPU / 1 GiB

On success it prints the live endpoints:

MCP endpoint : https://<app>.<region>.azurecontainerapps.io/mcp
Health check : https://<app>.<region>.azurecontainerapps.io/health

Set the Salesforce credentials on the running app:

az containerapp update --name ca-l2q-dev --resource-group rg-l2q-dev `
  --set-env-vars SF_AUTH_MODE=client_credentials `
                 SF_INSTANCE_URL=https://your-org.my.salesforce.com `
                 SF_CLIENT_ID=... SF_CLIENT_SECRET=...

Verify the deployment

$base = "https://<app>.<region>.azurecontainerapps.io"
# Health
Invoke-RestMethod "$base/health"          # -> {"status":"ok"}

# MCP handshake + tool list
$h = @{ "Content-Type"="application/json"; "Accept"="application/json, text/event-stream" }
Invoke-WebRequest "$base/mcp" -Method Post -Headers $h `
  -Body '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

A healthy server responds to initialize with serverInfo.name = "l2q" and serves the full Salesforce tool catalog (58 tools).

---

🛠 Tools Reference

Lead Management

Tool	Description
search_leads	Search by name, company, status or industry
get_lead	Full lead detail
get_recent_leads	All open leads (most recent first)
get_leads_by_status	Filter by status (New / Working / Qualified)

Account Management

Tool	Description
search_accounts	Search financial institution accounts
get_account	Full account detail
get_account_contacts	Contacts at an account
get_account_opportunities	Opportunities for an account

Contact Management

Tool	Description
search_contacts	Search by name, account or title
get_contact	Full contact profile
get_contact_activity_history	Call / email / meeting history

Opportunity Management

Tool	Description
search_opportunities	Filter by stage, value, account
get_opportunity	Full opportunity detail
get_open_opportunities	Open pipeline opportunities
get_opportunities_by_stage	Deals at a specific stage
get_pipeline_summary	Pipeline by stage with totals

Quote Management

Tool	Description
search_quotes	Find quotes by name or status
get_quote	Full quote with pricing
get_quote_pdf_package	Retrieve generated quote PDF artifacts/download URLs for email workflows
get_quote_products	Line items for a quote
get_opportunity_quotes	All quotes on an opportunity

Product Catalog

Tool	Description
search_products	Search the product catalog
get_product	Product details and list price

Activities & Tasks

Tool	Description
get_activity_timeline	Timeline for account/opp/contact
get_open_tasks	Open tasks sorted by priority

Create & Update Workflows (Draft-first)

Tool	Description
draft_create_record	Build a pre-create draft form for account/contact/case/opportunity/task/quote/lead/event
create_record	Create the record in Salesforce after explicit confirmation
get_record_for_update	Load a single record as an editable update draft
update_record	Apply updates in Salesforce after explicit confirmation
get_task / get_event / get_case	Retrieve single objects for update flows

Guided Entity Resolution & Lookups

Tool	Description
search_reference_records	Search accounts/contacts/opportunities/leads/cases for searchable picker dropdowns

Reporting

Tool	Description
run_salesforce_report	Run a Salesforce report and render tabular results in widget UI

Compliance & Deal Desk

Tool	Description
get_compliance_flags	MiFID II / GDPR / FCA flags
get_deal_desk_status	Approval process status

M365 Integration Helpers

Tool	Description
prepare_meeting_agenda	Teams meeting agenda from Salesforce data
prepare_email_summary	Professional email/proposal draft
get_sales_dashboard	Morning briefing: pipeline + tasks + leads
capture_lead_from_email	Build a lead draft from Outlook email context

Teams / Outlook / Calendar: Use the M365 Tooling Servers directly from your Copilot Declarative Agent for scheduling meetings and sending emails.

---

🎨 UI Widgets

HTML+Skybridge widgets are registered as MCP resources at ui://widgets/<name>.html:

Widget	Purpose
dashboard.html	Sales dashboard – pipeline + tasks + leads
leads.html	Lead list with status filter
opportunities.html	Kanban + table view
quotes.html	Quote cards + line items
accounts.html	Account cards + product catalog
contacts.html	Contact cards + detail profile
pipeline.html	Funnel + stacked bar chart
activities.html	Timeline + agenda/email viewer

Each widget reads structuredContent from the MCP _meta via the Skybridge postMessage protocol.

All widgets now render in compact mode by default (for inline chat iframes) and include a top-right Maximize/Restore button so the user can expand to full-height working mode as needed.

Create/update forms now support search-and-pick lookup flows for relationship IDs (AccountId, ContactId, WhoId, WhatId, OpportunityId, and opportunity PrimaryContactId) to avoid manual ID typing.

Prompt patterns for guided create flows

• Meeting-driven create: - I want to create an opportunity for the meeting with Jim that I just had - Agent flow: run search_reference_records to find the account/contact → open opportunity_form.html with prefilled data and candidate lookups.
• Blank create: - Create a new opportunity - Agent flow: run draft_create_record object_type=opportunity → use in-form pickers (Search/Pick) via search_reference_records to choose account/contact before create.

Prompt Library – L2Q lifecycle + advanced automation (with context expectations)

The prompts below are written for a Copilot Declarative Agent that can:

• read emails, Teams meeting summaries, and Teams messages,
• use this L2Q MCP server,
• use an external market-intelligence MCP for company/market context.

When a prompt references “this deal/account/lead”, expected resolution flow is:

1. Resolve entity from the current artifact context (email/thread/meeting),
2. If ambiguous, show candidates and ask user to pick,
3. Continue with deeper analysis/drafting.

6 prompts for L2Q phases + lifecycle reporting

Title	Prompt	Context source & expected flow
Lead Intake Snapshot	From recent client emails, Teams messages, and meeting summaries, produce a lead intake summary with qualification signals, urgency, and recommended next action.	Artifact-led: starts from current conversation artifacts; asks for entity confirmation only if multiple companies are referenced.
Qualification Decision	For lead [lead name or ID], assess qualify/nurture/disqualify and explain why using communication evidence plus external company and market context.	Entity-led: if no ID is provided, resolve by name and ask user to confirm best match.
Opportunity Readiness Check	For opportunity [name or ID], assess stage readiness, missing required data, stakeholder gaps, and what must be completed before stage progression.	Entity-led: if unspecified, shortlist active opportunities and let user choose one.
Quote Readiness Review	For opportunity [name or ID], evaluate quote readiness using recent client interactions and external market/company signals, then list blockers and next steps.	Entity-led: requires selected opportunity, then combines internal communication context + external intelligence.
Close Plan & Commit Confidence	For opportunity [name or ID], create a close plan with milestones, owners, confidence drivers, and top slippage risks.	Entity-led: one selected opportunity; returns plan and confidence narrative.
Weekly L2Q Report	Generate my weekly lifecycle report: phase distribution, phase movement, stalled deals, top risks, and next-week priorities.	Portfolio-led: runs across user scope; asks for region/team filter only if needed.

6 prompts for high-value, complex automation

Title	Prompt	Context source & expected flow
At-Risk Deal Rescue	Identify the most at-risk open opportunity in my portfolio and produce a two-week rescue plan using communication signals and external market/company context.	Portfolio-first then drill-down: proposes top at-risk deal, confirms, then generates rescue actions.
Executive Forecast Narrative	Build this week’s executive forecast narrative combining pipeline movement, communication evidence, and external market developments.	Portfolio-led synthesis for leadership reporting; no single-entity selection required.
Smart Meeting Prep Pack	For account [name], create a meeting prep pack with objective, talking points, likely objections, and market-aware responses based on latest interactions.	Account-led: resolves account, then pulls artifact context + external updates.
Post-Meeting Auto-Orchestration	From this meeting summary, generate follow-up actions, owner assignments, timeline updates, and a polished client follow-up message aligned to deal strategy.	Artifact-led: starts from current meeting summary; if multiple linked deals exist, asks user to pick target deal.
Renewal Risk & Expansion Plan	For account [name], create a renewal risk assessment and expansion plan using relationship signals, engagement patterns, and external outlook.	Account-led with multi-deal roll-up where applicable; returns 30/60/90 plan.
Board-Ready Pipeline Pack	Build a board-ready pipeline package for [portfolio/team/region] with confidence rationale, concentration risk, and recommended management interventions.	Portfolio-led: asks scope if not provided, then composes pack narrative and priorities.

Local widget preview with sample data

Use the preview script to render all widgets in a browser without running MCP tools:

scripts\run-widget-preview.ps1

Then open http://127.0.0.1:8090/ and select any widget page.

You can also choose a custom host/port:

scripts\run-widget-preview.ps1 -HostName 127.0.0.1 -Port 8091

---

🔐 Authentication Modes

Set SF_AUTH_MODE in your env file:

Mode	How it works
client_credentials	Server acquires a Salesforce token using SF_CLIENT_ID + SF_CLIENT_SECRET (Connected App)
oauth_bearer	Caller passes a Salesforce OAuth token in Authorization: Bearer … header – server forwards it
entra_sso	Caller passes a Microsoft Entra ID JWT – server validates it then forwards to Salesforce

---

🐳 Docker

docker build -t l2q-mcp .
docker run -p 8080:8080 \
  -e SF_AUTH_MODE=client_credentials \
  -e SF_INSTANCE_URL=https://your-org.my.salesforce.com \
  -e SF_CLIENT_ID=... \
  -e SF_CLIENT_SECRET=... \
  l2q-mcp

The server exposes a /health endpoint at GET /health that returns {"status": "ok"} — suitable for container orchestration liveness/readiness probes.

---

🧪 Testing

pip install -e .
python -m pytest tests/ -v

---

📋 M365 Copilot Declarative Agent Setup

1. Create a new Declarative Agent in VS Code using M365 Agents Toolkit
2. Add the RemoteMCP plugin with endpoint https://<your-server>/mcp
3. Add the M365 Tooling Servers for Teams, Outlook and Calendar
4. (Optional) Add a market-intelligence MCP plugin to the same agent for external company/market context
5. Set Authorization: Bearer <Entra token> in the RemoteMCP plugin config

---

🤝 Copilot Cowork Plugin

The cowork/ folder contains a Copilot Cowork plugin that reuses this same MCP server as a remoteMcpServer connector. Where the Declarative Agent is the chat surface for fast lookups, the Cowork plugin is the co-work surface for deep work — it produces finished email, Word, PowerPoint and Excel artifacts. It ships five skills across two fintech personas:

Skill	Persona	Built-in surfaces
account-research-brief	Account executive	Word, Email
deal-followup-automation	Account executive	Email, Word
pitch-deck-builder	Account executive	PowerPoint
client-meeting-scheduler	Account executive	Calendar, Email
pipeline-review-pack	Sales manager	Excel, PowerPoint, Email

Package it with cowork/scripts/package.ps1 and sideload sales-assistant-cowork.zip. See cowork/README.md for the full story, connector auth, and packaging steps.

---

🏢 About

Sales Assistant is a Microsoft 365 Copilot extensibility sample for fintech B2B sales — built for professionals managing large, complex deals with financial-institution clients in a highly regulated environment (MiFID II / GDPR / FCA). It demonstrates an MCP server powering both a Declarative Agent and a Copilot Cowork plugin over live Salesforce data.