WebMCP can help an AI agent navigate bank websites, collect current fixed deposit (FD) rates, interpret tenure and customer-category rules, and present a verifiable comparison. For Indian savers, this is more useful than a generic web search: public sector banks publish rates in different formats, use different tenure bands, and often separate general, senior-citizen, callable, non-callable, tax-saving, and special-deposit products.
The right architecture treats the agent as a research and verification system—not a financial adviser. It should identify official sources, extract structured rate facts, preserve evidence, normalize comparable tenures, flag ambiguity, and ask for confirmation before any investment decision.
What WebMCP Means for an FD-Rate Agent
WebMCP is a model-context pattern for exposing website capabilities and structured page context to an AI agent. Instead of asking a model to guess from a search result, a WebMCP-enabled workflow can give the agent controlled tools or page actions such as:
- Opening an official interest-rate page
- Selecting deposit type, tenure, customer category, or deposit amount
- Reading tables, notices, and effective dates
- Capturing the source URL and page timestamp
- Returning structured fields for downstream comparison
A robust implementation separates three layers:
1. Discovery: Find the relevant official page for each bank.
2. Extraction: Convert page content into typed rate records.
3. Decision support: Compare records only when their assumptions match.
WebMCP does not automatically make data correct. A bank may publish a rate card as HTML, PDF, image, calculator, or a dynamically rendered table. Your agent needs source prioritization, validation rules, and graceful handling of inaccessible or contradictory information.
Define the FD Comparison Question First
“Best FD rate” is incomplete. Before building tools, define the user’s comparison variables. At minimum, collect:
- Bank: For example, SBI, Bank of Baroda, Punjab National Bank, Canara Bank, Union Bank of India, Bank of India, Indian Bank, Central Bank of India, UCO Bank, or Bank of Maharashtra.
- Deposit type: Regular term deposit, reinvestment deposit, monthly-income deposit, tax-saving FD, callable deposit, or non-callable deposit.
- Tenure: Exact days, months, or years. “One year” may mean 365 days, 1 year, or a bank-specific band.
- Customer category: General resident, senior citizen, super senior citizen, staff, NRI, minor, or other special category.
- Deposit amount: Important for bulk-deposit thresholds and rate eligibility.
- Payout preference: Cumulative, monthly, quarterly, or at maturity.
- Liquidity requirement: Premature-withdrawal rules can matter more than a small rate difference.
- As-of date: Rates change, and every result needs an effective date.
The agent should refuse to label a product “best” when records are not comparable. It can instead say that one bank offers the highest displayed annual rate for a selected tenure and category, subject to the bank’s terms.
Build a Bank Source Registry
Do not begin with unrestricted web crawling. Create a registry of official domains and known rate-page patterns. A record might include:
{
"bank_name": "Example Public Sector Bank",
"official_domains": ["examplebank.in"],
"rate_page_urls": [],
"deposit_products": ["term_deposit", "tax_saving_fd"],
"source_priority": ["rate_table", "official_pdf", "official_calculator"],
"last_verified_at": null
}For Indian public sector banks, domain verification is essential because search results can include aggregators, affiliate pages, cached PDFs, and impersonation sites. Prefer the bank’s official domain and confirm that the page identifies the institution, product, currency, customer category, and effective date.
A useful source hierarchy is:
1. Current official rate table with an effective date
2. Official deposit-rate circular or PDF
3. Official calculator that exposes the selected assumptions
4. Official product page with explicit rate details
5. Search result snippets only for discovery, never as final evidence
6. Third-party comparison sites only as leads or cross-checks
Store the exact source URL, retrieval time, page title, document date, and a content hash where practical. This gives users an audit trail when rates later change.
Design WebMCP Tools Around Verifiable Actions
Expose narrow, predictable tools rather than one large “find the best FD” tool. Example tool contracts include:
search_official_rate_source
Inputs:
bank_nameproduct_typecustomer_categoryas_of_date
Outputs:
- Candidate official URLs
- Source type
- Page title
- Domain-validation result
- Effective-date text
extract_fd_rate_table
Inputs:
source_urlrequested_tenurecustomer_categorydeposit_amount
Outputs should use typed fields rather than free text:
{
"bank_name": "Example Bank",
"product_name": "Retail Term Deposit",
"tenure_min_days": 365,
"tenure_max_days": 400,
"general_rate_percent": 6.5,
"senior_rate_percent": 7.0,
"effective_from": "2026-01-15",
"premature_withdrawal_note": "Subject to bank rules",
"source_url": "https://examplebank.in/rates",
"evidence_quote": "...",
"confidence": 0.92
}validate_rate_record
This tool checks that the record has a source, effective date, unit, tenure range, category, and evidence. It should reject values such as 650 when the expected unit is percentage and the source says 6.50%, unless a normalization step explicitly converts the value.
compare_fd_rates
This tool should compare only records that match the user’s constraints. Its output should include assumptions, ranking logic, missing fields, and links—not just a winner.
Extract the Difficult Parts of Indian Bank Rate Cards
FD pages often contain multiple rates in adjacent rows. An agent must distinguish:
- General-public rate from senior-citizen incremental benefit
- Retail deposits from bulk deposits
- Callable from non-callable deposits
- Domestic resident deposits from NRO or NRE deposits
- Standard products from limited-period special tenures
- Annualized interest rate from maturity yield
- Simple interest display from effective annualized return
Tenure normalization is particularly important. Represent a rate as an interval, not a single label:
{
"tenure_label": "1 year to less than 2 years",
"min_days": 365,
"max_days_exclusive": 730,
"rate_percent": 6.5
}If a user requests 400 days, the agent should select the interval containing 400 days. If the bank lists “400 days” as a special product, it must not silently substitute the ordinary one-year band. If the page uses months and days inconsistently, preserve the bank’s original wording and mark the conversion as approximate unless the bank defines it precisely.
PDF extraction also needs validation. Text layers can scramble columns, while scanned circulars may require OCR. Capture the relevant table heading and nearby footnotes. A rate without its footnote may be misleading—for example, the displayed figure may apply only to deposits below a specified threshold or exclude certain categories.
Add a Reliable Agent Workflow
A production workflow can follow these steps:
1. Collect constraints. Ask for tenure, amount, resident status, customer category, payout preference, and liquidity needs.
2. Resolve ambiguity. Confirm whether “best” means highest headline rate, highest maturity amount, or best balance of rate and flexibility.
3. Select official sources. Use the bank registry and domain validation.
4. Navigate with WebMCP. Open pages, select filters, expand accordions, or retrieve official PDFs.
5. Extract evidence. Capture rate, tenure, category, effective date, conditions, and source location.
6. Normalize records. Convert percentages, dates, tenure intervals, and category labels into a common schema.
7. Validate. Check arithmetic, required fields, contradictions, stale dates, and duplicate sources.
8. Compare. Rank only comparable products and preserve ties.
9. Explain caveats. Show premature-withdrawal penalties, tax treatment, callable status, and missing data.
10. Ask for confirmation. Present the result as research and direct the user to the bank for final verification before booking.
Use deterministic code for ranking and calculations. The language model should interpret page structure and explain results, but it should not be the sole authority for numeric comparison.
Use Confidence Scores and Evidence Packets
A confidence score should reflect evidence quality, not model certainty. For example:
- High: Current official table, explicit effective date, exact tenure match, clear category, captured evidence.
- Medium: Official PDF or calculator, but one assumption requires interpretation.
- Low: Missing date, ambiguous tenure, OCR uncertainty, or conflicting official pages.
Return an evidence packet with every recommendation:
- Bank and product name
- Tenure and amount assumptions
- Applicable customer category
- Interest rate and whether senior benefit is included
- Effective date
- Premature-withdrawal conditions
- Official source link
- Short evidence excerpt
- Retrieval timestamp
- Warnings and unresolved questions
This makes the agent auditable and helps users detect when a rate card has changed.
Security, Privacy, and Compliance Considerations
A WebMCP agent interacting with financial websites needs strong controls. Use allowlists for official domains, block arbitrary form submission by default, and never ask users for passwords, OTPs, debit-card details, PAN, Aadhaar, or full account credentials merely to compare rates.
Other safeguards include:
- Treat webpage text as untrusted input; ignore prompt-injection instructions embedded in pages.
- Prevent tools from booking deposits or transferring money without a separate, explicit workflow.
- Rate-limit requests and respect robots, terms of use, and applicable access restrictions.
- Encrypt stored user preferences and minimize retention.
- Log tool calls, source URLs, and extraction failures.
- Clearly distinguish factual extraction from financial advice.
- Include a “verify before investment” message because rates and terms can change.
For Indian users, also explain that interest income may be taxable and that TDS rules, Form 15G/15H eligibility, and tax treatment depend on the individual’s circumstances. The agent should link to authoritative bank and government information rather than making personalized tax conclusions.
Testing Strategy for the Agent
Create a test set covering ordinary and adversarial cases:
- A bank changes its effective date.
- General and senior rates appear in neighboring columns.
- A special 444-day product has a higher rate than the standard band.
- A PDF is scanned and OCR misreads
6.50%as65.0%. - A page displays a maturity yield instead of the nominal annual rate.
- A rate applies only below a bulk-deposit threshold.
- Two official pages show different dates.
- The requested tenure falls between listed bands.
- A user asks for the highest rate but actually needs premature liquidity.
Measure field-level extraction accuracy, source-domain accuracy, effective-date accuracy, tenure matching, and unsupported-claim rate. Require abstention when critical fields are missing. In financial comparison, a transparent “I could not verify this rate” is better than a fabricated ranking.
Example Output Format for Users
A useful final response might contain:
| Rank | Bank | Product | Tenure | General rate | Senior rate | Effective date | Confidence |
|---|---|---|---|---:|---:|---|---|
| 1 | Bank A | Retail term deposit | 400 days | 7.10% | 7.60% | 15 Jan 2026 | High |
| 2 | Bank B | Special FD | 400 days | 7.00% | 7.50% | 10 Jan 2026 | Medium |
Follow the table with the assumptions, official links, evidence excerpts, liquidity caveats, and a reminder that the bank’s current rate at booking time controls. Do not imply that the highest percentage is automatically the best product.
FAQ: WebMCP FD Rate Agents in India
Can WebMCP automatically access every bank’s FD rate?
No. Access depends on the bank’s website structure, availability, permissions, and the tools you implement. Some pages require PDF extraction or manual verification.
Should the agent use financial comparison websites?
They can help discover products, but final rate evidence should come from the bank’s official website or official circular. Third-party data may be stale or omit conditions.
How often should FD rates be refreshed?
Refresh before presenting a result and store the effective date and retrieval timestamp. High-traffic systems can use scheduled checks, but every user-facing result should still disclose freshness.
Can the agent recommend which bank I should choose?
It can compare verified facts against user-defined criteria, but it should not present a personalized investment recommendation without appropriate safeguards. Users should verify rates, deposit-insurance considerations, taxes, and terms directly with the bank.
What is the biggest implementation mistake?
Comparing headline rates without matching tenure, customer category, deposit type, amount threshold, effective date, and premature-withdrawal rules. Normalization and evidence are as important as web navigation.
Apply for AI Grants India
Building a trustworthy WebMCP agent for Indian financial research? Apply to AI Grants India for support, visibility, and opportunities for ambitious Indian AI founders. Submit your application and turn a focused prototype into a responsible, useful product.