Part 3: Add custom tools (FRED MCP + ThinQ MCP)

Part 3: Add custom tools (FRED MCP + ThinQ MCP)

Previous step: Part 2: Add pre-built tools

Back to start: Overview

---

In this part, you will add two custom MCP servers that are not provided by Microsoft by default.

• FRED Economic MCP — U.S. Federal Reserve economic data
• ThinQ MCP — query and control LG IoT appliances

Both servers are private MCP servers deployed to Azure Container Apps. You connect by entering the URL and API key directly.

---

1. How to add an MCP tool (common steps)

On the agent overview screen, click the Tools section → + Add a tool.

In the add-tool popup, select the MCP tab → the Model Context Protocol (MCP) option at the bottom.

Pre-built vs Custom: The Work IQ tools added in Part 2 were selected from the catalog, but for custom MCP you enter the URL and authentication information directly.

---

2. Connect the FRED Economic MCP server

2-1. Introduction to FRED MCP

FRED (Federal Reserve Economic Data) is a database of more than 810,000 economic time series provided by the Federal Reserve Bank of St. Louis. Through this MCP server, the agent can query and analyze economic indicators in real time.

Item	Details
Data source	FRED (Federal Reserve Economic Data)
Number of series	810,000+ economic time series
Provided tools	6 theme tools + 1 search tool
Transport	Streamable HTTP
Deployment environment	Azure Container Apps (Korea Central)

2-2. Enter the server URL and authentication information

Enter the information below exactly:

Item	Value
Server name	FRED_Economic_MCP
Server description	An MCP server for querying U.S. Federal Reserve (FRED) economic data. It provides economic indicators across six themes: consumer demand, commodities and exchange rates, macroeconomy, industrial production, EV and energy markets, and robotics and actuators. It can also search more than 810,000 time series by keyword.
Server URL	https://fred-mcp-python.purplehill-ba85d6d0.koreacentral.azurecontainerapps.io/mcp
Authentication	API key
Authentication type	Header
Header name	x-fred-api-key
API key value	(Provided separately during the lab)

After entering the information, click Create.

Why the server description matters: Copilot Studio’s generative AI orchestrator uses the server description to decide when to call this tool. The description must be specific enough for the agent to behave correctly.

2-3. Check the tool list

When the connection succeeds, the seven tools below are displayed:

Tool name	Description	Key series
get-consumer-demand	Consumer demand indicators	PCEDG (durable goods consumption), UMCSENT (consumer sentiment), HOUST (housing starts), HSN1F (home sales), TOTALSA (auto sales)
get-cost-pressure	Cost/margin pressure indicators	DEXKOUS (KRW/USD exchange rate), DCOILWTICO (WTI oil price), WPUSI019011 (copper PPI), WPU101 (steel PPI), PCU325211325211 (synthetic resin PPI)
get-macro-environment	Macroeconomic environment	GDPC1 (real GDP), CPIAUCSL (CPI), CUSR0000SAD (durable goods CPI), FEDFUNDS (federal funds rate), UNRATE (unemployment rate), DGS10 (10-year Treasury), DGS2 (2-year Treasury)
get-industry-production	Industry/manufacturing trends	INDPRO (industrial production index), DGORDER (durable goods orders), AMTMNO (manufacturing orders), IPMAN (manufacturing production)
get-ev-energy-market	EV/energy market	TOTALSA (auto sales), AISRSA (inventory-to-sales ratio), DCOILWTICO (oil price), GASREGW (retail gasoline price), IPG3361T3S (auto production)
get-robotics-actuator	Robotics & actuators	NASDAQNQROBO (AI Robotics Global), NASDAQNQROBOUS (AI Robotics US), PCU333995333995 (actuator PPI), WPU114303 (actuator and parts PPI), IPG3332S (industrial machinery production), A33ENO (industrial machinery orders), IPG3336S (power transmission equipment production), ATGPNO (power transmission equipment orders)
search-fred-series	Search FRED series	Explore 810,000+ time series by keyword

2-4. Tool parameters

All theme tools (get-*) use the same parameters:

Parameter	Type	Default	Description
indicator	string (optional)	Query all	Query only a specific series ID (for example, FEDFUNDS)
period	enum	1y	Query period: 6m, 1y, 3y, 5y
units	enum	lin	Data transformation: lin (raw), chg (change), pch (% change), pc1 (year-over-year %), pca (annualized), log (natural log)

Parameters for the search-fred-series tool:

Parameter	Type	Default	Description
query	string (required)	—	Search keyword (for example, semiconductor production)
limit	integer	10	Number of search results (1-50)

Lab test: In the agent test pane, enter “Show me the U.S. unemployment-rate trend over the last 3 years.” The agent should call get-macro-environment(indicator="UNRATE", period="3y").

---

3. Connect the ThinQ MCP server

3-1. Introduction to ThinQ MCP

ThinQ MCP is an MCP server that queries and controls IoT appliances registered in the LG ThinQ platform.

Item	Details
Target devices	Air conditioners, washing machines, refrigerators, air purifiers, dryers, and more
Provided features	Device query, status check, control commands, energy usage query
Transport	Streamable HTTP
Deployment environment	Azure Container Apps (Korea Central)

3-2. Enter the server URL and authentication information

Select + Add a tool → MCP → Model Context Protocol (MCP) again, and enter the information below:

Item	Value
Server name	ThinQ_MCP
Server description	An MCP server for querying and controlling smart appliances on the LG ThinQ IoT platform. It provides features for checking the status of air conditioners, washing machines, refrigerators, and other devices, executing control commands, and querying energy usage.
Server URL	https://thinqmcp-http-typescript.kindmoss-12649bc0.koreacentral.azurecontainerapps.io/mcp
Authentication	API key
Authentication type	Header
Header name	x-pat-token
API key value	(Provided separately during the lab)

After entering the information, click Create.

3-3. Check the tool list

When the connection succeeds, the tools below are displayed:

Tool name	Description
getDevices	Query the list of all registered devices
getDeviceStatus	Query the current status of a specific device (operation mode, temperature, whether it is running, and so on)
controlDevice	Execute device control commands (change air conditioner temperature, start a washing machine, and so on)
getEnergyUsage	Query energy usage by device

3-4. Tool parameters

getDeviceStatus / controlDevice / getEnergyUsage:

Parameter	Type	Description
deviceId	string (required)	Unique device ID (queried with getDevices)

Additional parameters for controlDevice:

Parameter	Type	Description
command	string (required)	Control command (for example, setTemperature, start, stop)
value	string/number	Value for the command (for example, temperature 24)

Lab test: Enter “Show me the list of registered appliances.” The agent calls getDevices.

---

4. Check all tools

After all tools are added, you should see the following in the agent’s Tools section:

Tool group	Number of tools	Type
Work IQ Mail MCP	4-5	Pre-built
Work IQ Teams MCP	3-4	Pre-built
Work IQ Copilot MCP	1-2	Pre-built
FRED Economic MCP	7	Custom
ThinQ MCP	4	Custom

---

5. Integration tests

5-1. Single-tool tests — check basic behavior

In the agent test chat, first use simple prompts to confirm that each tool is called correctly:

FRED macroeconomy (get-macro-environment):

#	Test question	Expected parameters
1	“Show me the U.S. unemployment-rate trend.”	indicator=UNRATE, period=1y
2	“Show me the 3-year CPI trend as a year-over-year change rate.”	indicator=CPIAUCSL, period=3y, units=pc1
3	“Query 5-year federal funds rate data as raw values.”	indicator=FEDFUNDS, period=5y, units=lin
4	“Compare 10-year and 2-year Treasury yields and check whether the yield curve is inverted.”	Query all without indicator → compare DGS10, DGS2
5	“Show me the 3-year trend for durable goods CPI (CUSR0000SAD). Analyze whether monitor and display product prices are rising.”	indicator=CUSR0000SAD, period=3y

FRED consumer demand (get-consumer-demand):

#	Test question	Expected parameters
6	“Show me U.S. durable goods consumption expenditures for the last 1 year.”	indicator=PCEDG, period=1y
7	“Show me the 3-year trend for the consumer sentiment index as a year-over-year change rate.”	indicator=UMCSENT, period=3y, units=pc1
8	“Are new housing starts increasing? Tell me using the latest 6 months of data.”	indicator=HOUST, period=6m
9	“Show me the year-over-year change rate for total U.S. auto sales.”	indicator=TOTALSA, units=pc1

FRED cost pressure (get-cost-pressure):

#	Test question	Expected parameters
10	“Tell me the KRW/USD exchange-rate trend over the last 6 months.”	indicator=DEXKOUS, period=6m
11	“How has the WTI crude oil price moved recently?”	indicator=DCOILWTICO
12	“Compare copper PPI and steel PPI.”	Query all without indicator
13	“Show me the synthetic resin PPI trend. Analyze how monitor housing and plastic component costs are changing.”	indicator=PCU325211325211

FRED industry/manufacturing (get-industry-production):

#	Test question	Expected parameters
14	“Query the 3-year U.S. industrial production index trend as a year-over-year change rate.”	indicator=INDPRO, period=3y, units=pc1
15	“Are new durable goods orders increasing? Analyze whether B2B orders related to IT equipment and displays are recovering.”	indicator=DGORDER
16	“Analyze factory operating conditions by combining the manufacturing production index and new orders.”	Query all without indicator

FRED EV/energy (get-ev-energy-market):

#	Test question	Expected parameters
17	“What is the auto inventory-to-sales ratio? Are inventories building up?”	indicator=AISRSA
18	“Show me WTI oil prices and retail gasoline prices together.”	Query all without indicator

FRED series search (search-fred-series):

#	Test question	Expected keyword
19	“Search for FRED series related to semiconductor production.”	semiconductor production
20	“Search for series related to computer and electronic product manufacturing.”	computer electronic product manufacturing
21	“Search whether FRED has data related to display panel or flat panel.”	display panel / flat panel
22	“Find indicators related to information processing equipment.”	information processing equipment

FRED robotics & actuators (get-robotics-actuator):

#	Test question	Expected parameters
23	“Query all economic indicators related to robotics and actuators.”	Query all without indicator
24	“Show me the Nasdaq AI Robotics Index trend over the last 3 years.”	indicator=NASDAQNQROBO, period=3y
25	“How has actuator PPI changed recently? Judge whether demand is increasing.”	indicator=PCU333995333995
26	“Show me industrial machinery new orders and the production index. Check whether demand for robotic equipment is recovering.”	Query all without indicator → check IPG3332S, A33ENO
27	“Show me 5-year power transmission equipment new orders data as a year-over-year change rate. I want to understand the demand trend for servo motors and reducers.”	indicator=ATGPNO, period=5y, units=pc1
28	“Show me the year-over-year change rate for fluid power cylinder and actuator PPI.”	indicator=WPU114303, units=pc1

Work IQ (Mail/Teams/Copilot):

#	Test question	Expected called tool
29	“Summarize emails related to ‘market analysis’ that I received in the last week.”	Work IQ Mail
30	“Search recent Teams conversations with the keyword ‘display costs’.”	Work IQ Teams
31	“Find whether there are consulting reports or market reports among my recently received emails.”	Work IQ Mail

ThinQ (IoT):

#	Test question	Expected called tool
32	“Show me the list of registered appliances.”	ThinQ getDevices

---

5-2. FRED complex analysis tests — multiple-tool combinations

Check whether multiple FRED tools are called sequentially from a single question. The tests are sorted by difficulty (⭐).

⭐ Two-tool combinations (basic):

#	Test question	Expected call order
1	“Analyze exchange rates and oil prices together and tell me the impact on import costs.”	get-cost-pressure (DEXKOUS + DCOILWTICO) → synthesize
2	“Evaluate U.S. economic conditions by combining GDP and the unemployment rate.”	get-macro-environment (GDPC1 + UNRATE) → synthesize
3	“Look at copper and synthetic resin PPI together with exchange rates, and diagnose the level of import cost pressure for display components.”	get-cost-pressure (DEXKOUS + WPUSI019011 + PCU325211325211) → synthesize
4	“Compare durable goods CPI with durable goods consumption expenditures, and analyze whether consumption is increasing relative to monitor and display prices.”	get-macro-environment (CUSR0000SAD) + get-consumer-demand (PCEDG) → compare
5	“Combine the consumer sentiment index and durable goods consumption to analyze the consumption outlook for IT electronics (monitors, TVs).”	get-consumer-demand (UMCSENT + PCEDG) → outlook
6	“Analyze the AI Robotics Index and actuator PPI together to compare market sentiment and real demand in the Physical AI market.”	get-robotics-actuator (NASDAQNQROBO + PCU333995333995) → compare
7	“Combine industrial machinery new orders and actuator PPI to judge whether this is the time to expand investment in robotic automation equipment.”	get-robotics-actuator (A33ENO + PCU333995333995) → investment judgment

⭐⭐ Three-tool combinations (intermediate):

#	Test question	Expected call order
6	“Analyze the display margin outlook by reviewing cost trends from exchange rates, copper, and synthetic resin PPI, then checking price pass-through with durable goods CPI.”	get-cost-pressure → get-macro-environment (CUSR0000SAD) → margin analysis
7	“Check yield-curve inversion, GDP, and unemployment, then add industrial production index and durable goods orders trends to evaluate recession risk and manufacturing impact.”	get-macro-environment → get-industry-production → comprehensive judgment
8	“Combine housing starts, consumer sentiment, and durable goods consumption expenditures to analyze the demand outlook for home monitors and TV displays.”	get-consumer-demand (HOUST + UMCSENT + PCEDG) → demand outlook
9	“Check the interest-rate and GDP environment, then add durable goods orders trends to diagnose the B2B display equipment investment environment.”	get-macro-environment → get-industry-production (DGORDER) → B2B investment judgment
10	“Combine robotics and actuator indicators with industrial production and macroeconomic data to evaluate the industrial robot equipment investment environment.”	get-robotics-actuator → get-industry-production → get-macro-environment → comprehensive evaluation
11	“Analyze power transmission equipment new orders, the AI Robotics Index, and interest rates together to diagnose the servo motor and reducer market outlook.”	get-robotics-actuator (ATGPNO + NASDAQNQROBO) → get-macro-environment (FEDFUNDS) → outlook

⭐⭐⭐ Four or more tool combinations (advanced):

#	Test question	Expected call order
10	“Review display cost trends using exchange rates, copper, and synthetic resin PPI, then analyze durable goods CPI and consumer sentiment together to evaluate the monitor market profitability outlook.”	get-cost-pressure → get-macro-environment → get-consumer-demand → comprehensive profitability
11	“Combine yield-curve inversion, GDP, unemployment, and the industrial production index, then add exchange rates and commodity PPI to evaluate this year’s display manufacturing investment environment.”	get-macro-environment → get-industry-production → get-cost-pressure → comprehensive judgment
12	“After searching for computer and electronic product manufacturing series, combine the related indicators with exchange-rate and commodity data and durable goods CPI to comprehensively diagnose display industry conditions.”	search-fred-series → get-cost-pressure → get-macro-environment → comprehensive diagnosis
13	“Query all robotics and actuator indicators, add exchange rates, commodity PPI, interest rates, and industrial production data, and create a briefing that comprehensively evaluates the Physical AI business investment environment.”	get-robotics-actuator → get-cost-pressure → get-macro-environment → get-industry-production → comprehensive briefing

---

5-3. FRED + Work IQ complex analysis tests — macroeconomy × internal communications

These scenarios combine FRED economic data with email, Teams conversations, and consulting materials for practical business judgment.

⭐ Simple — FRED + email/chat combinations:

#	Test question	Expected behavior
1	“Summarize recently received emails related to ‘market analysis,’ and also show U.S. unemployment-rate and GDP trends.”	Work IQ Mail → get-macro-environment → provide information in parallel
2	“Search recent Teams conversations related to ‘commodity prices,’ and show actual copper, steel, and synthetic resin PPI trends together.”	Work IQ Teams → get-cost-pressure → compare conversation content vs actual data
3	“Find content in recent emails that includes the keyword ‘exchange rate,’ and also query KRW/USD exchange-rate data for the last 6 months.”	Work IQ Mail → get-cost-pressure (DEXKOUS) → compare email content with actual trends

⭐⭐ Intermediate — internal-material validation + economic-data cross-analysis:

#	Test question	Expected behavior
4	“Find recently received consulting report emails, summarize the key points, and validate the economic indicators mentioned in the reports with actual data from FRED.”	Work IQ Mail (consulting email search) → related FRED tools → cross-validation
5	“Search recent Teams discussions related to the ‘display market,’ and fact-check the team discussion using durable goods consumption, consumer sentiment, and durable goods CPI data.”	Work IQ Teams → get-consumer-demand + get-macro-environment → fact check
6	“Reevaluate the pricing policy based on commodity price changes. Find related internal emails and Teams conversations, and strengthen the rationale with FRED exchange-rate and commodity PPI data.”	Work IQ Mail + Teams → get-cost-pressure → pricing policy review
7	“Find and summarize recent email reports related to the ‘North America market,’ then add GDP, unemployment-rate, and interest-rate data to evaluate whether the report content matches current economic conditions.”	Work IQ Mail → get-macro-environment → compare report vs actual economy

⭐⭐⭐ Advanced — comprehensive judgment + communication deliverables:

#	Test question	Expected behavior
8	“Summarize recent reporting emails, query exchange rates, commodity PPI, durable goods CPI, and the industrial production index, compare and analyze them against market conditions, then summarize the key risk points and share them to Teams as a card.”	Work IQ Mail → get-cost-pressure + get-macro-environment + get-industry-production → risk analysis → Teams card
9	“Search Teams discussions related to ‘rising display costs,’ analyze the level of cost pressure using copper and synthetic resin PPI and exchange-rate data, then write an opinion email with response options.”	Work IQ Teams → get-cost-pressure → analysis → Work IQ Mail (write email)
10	“Comprehensively search internal consulting-material emails and Teams discussions, query GDP, interest rates, durable goods orders, and exchange-rate data from FRED, and write a briefing email on this year’s display business unit investment environment.”	Work IQ Mail + Teams → get-macro-environment + get-industry-production + get-cost-pressure → write briefing email
11	“Write a decision-opinion email about entering a new market. Find related internal materials in email and Teams, then combine GDP, interest rates, consumer sentiment, industrial production index, and exchange-rate data to build the rationale.”	Work IQ Mail + Teams → get-macro-environment + get-consumer-demand + get-industry-production + get-cost-pressure → write decision email
12	“Search FRED for computer and electronic product manufacturing series, summarize related Teams discussions and email reports together, and share a display-industry business outlook report as a Teams card.”	search-fred-series → Work IQ Teams + Mail → comprehensive report → Teams card
13	“Search Teams discussions related to ‘robotics’ or ‘automation,’ combine robotics and actuator indicators with industrial production and interest-rate data, and write a briefing email on the Physical AI business investment environment.”	Work IQ Teams → get-robotics-actuator + get-industry-production + get-macro-environment → write briefing email

---

5-4. Complex scenarios — full workflow tests

Check whether the business scenarios defined in the agent’s Instructions work end to end:

Test question	Expected behavior
“Review the North America market strategy.”	SharePoint search → FRED query → gap analysis → Teams adaptive card
“Reevaluate the pricing policy based on commodity price changes.”	SharePoint search → FRED cost indicators → margin analysis → Teams notification
“Summarize recent reporting emails and compare them with market conditions.”	Mail query → FRED validation → risk analysis → Teams notification
“Write a decision-opinion email about entering a new market.”	FRED + SharePoint → decision email draft → Teams notification
“Judge whether this is the right time to expand investment in robotics and automation equipment.”	FRED Tool 6 (robotics and actuators) + Tool 3 (interest rates) + Tool 4 (industrial production) → investment judgment
“Create a briefing on the investment environment for the Physical AI business.”	FRED Tool 6 + Tool 2 (exchange rates and commodities) + Tool 3 (macroeconomy) → briefing → Teams card

---

✅ Part 3 completion checklist

• Has FRED Economic MCP been connected, and are seven tools enabled?
• Has ThinQ MCP been connected, and are four tools enabled?
• Do all single-tool tests pass?
• Are multiple tools called sequentially in complex scenario tests?
• Are Teams adaptive card notifications sent correctly?

In the next step, you will add an email-received trigger. ➡️ Part 4: Add an email-received trigger