# Changelog

> Notable changes to the Portfolio Intel API, MCP server, and docs.

## 2026-10-17 — Expense ratios from AMFI TER disclosures

- `portfolio-health-check` 1.5.0: enrichment sets each fund's expense ratio from AMFI's TER disclosures (Direct or Regular, matching the holding's plan), so the expense ratio check runs. Its `source` is now `amfi_ter`, and it reports `evaluated_count` and `eligible_count`. When it had a TER for only some funds, the headline ends "(expense ratio checked for N of M funds)" and the `run_skill` text output lists it under "Partly evaluated". A flag's `reasoning` names the date of the TER disclosure.
- With enrichment off, the expense ratio check's `skipped_reason` says it needs expense ratios (TER) from enrichment. The underperformance check keeps its reason about fund returns and benchmarks.
- The MCP App shows "Checked N of M funds" on the expense ratio step and lists "AMFI TER disclosures" as a source.

## 2026-10-17 — Expense ratio check reported as skipped

- `portfolio-health-check` 1.4.0: when no fund has an expense ratio (TER), the expense ratio check is `skipped` with a `skipped_reason`, listed in `checks_skipped`, and the headline ends "(partial — no expense ratio data)". Earlier versions reported it as passed. Enrichment does not fetch TER yet, so the check is skipped on every run for now.
- The hosted `run_skill` text output lists skipped checks under "Not evaluated". The npm server does the same from its next release. See [Skills](https://docs.portfoliointel.co.in/skills#understanding-the-output).

## 2026-10-15 — Health check output is analytics only

- `portfolio-health-check` 1.3.0: each flag's `suggested_action` now explains what the flag means instead of suggesting a trade, and the result carries a `disclaimer`. The expense-ratio flag states the 1.5% threshold it fires at and no longer points Direct-plan holders to a Direct plan. See [Skills](https://docs.portfoliointel.co.in/skills#understanding-the-output).
- The four MCP prompts ask for findings with numbers, not buy, sell or switch recommendations. `tax_planning_india` states the specified-fund rule correctly (more than 65% in debt and money market instruments) and covers grandfathering and fund-of-funds holding periods.
- `compare_snapshots` is no longer listed in the capability manifest or the `tools-claude.json` and `tools-openai.json` definitions.
- `server.json` now passes the MCP registry schema, under the `in.co.portfoliointel` namespace.

## 2026-09-29 — Docs: agents and getting a CAS

- New guide: [Get your CAS](https://docs.portfoliointel.co.in/guides/get-your-cas). Covers requesting a statement, Gmail import in the mobile app, and how an agent with its own Gmail access can fetch and parse the statement.
- [MCP tools](https://docs.portfoliointel.co.in/mcp/tools) now shows which tools the hosted and npm servers each expose, marks `compare_snapshots` as coming soon, and documents the four MCP prompts.
- [MCP authentication](https://docs.portfoliointel.co.in/mcp/authentication) documents token lifetimes (1-hour access token, rotating 30-day refresh token) and corrects the scopes table.

## 2026-09-26 — Free key limits

- Unclaimed keys from [`POST /api/v1/register`](https://docs.portfoliointel.co.in/api-reference/register) and the OAuth consent page get 30 credits for 7 days. Claiming a key raises it to 100 credits and 30 days from the claim.
- New `429 DAILY_CAP_REACHED` when the day's free keys are used up.
- Per-network limits count an IPv6 /64 as one network.

## 2026-09 — Parse on your own device

- **In-browser terminal** on [portfoliointel.co.in](https://portfoliointel.co.in/#see-it-work): parse a CAS in the tab, nothing uploaded. See [Parse on your own device](https://docs.portfoliointel.co.in/local/overview).
- **`portfolio-intel` CLI** (built, not yet on npm): the same commands offline, plus `parse --json` with exit codes for scripts. See [CLI](https://docs.portfoliointel.co.in/local/cli).

## 2026-09-21 — Parse limits and clearer errors

- `POST /api/v1/parse` now enforces 10 MB, 150 pages and a 45-second budget, returning `413 FILE_TOO_LARGE`, `413 TOO_MANY_PAGES` and `422 PARSE_TIMEOUT`. Parsing runs in isolated workers; `503 SERVER_BUSY` means retry shortly. See [Limits](https://docs.portfoliointel.co.in/api-reference/parse#limits) and [Errors](https://docs.portfoliointel.co.in/api-reference/parse#errors).
- [Upload sessions](https://docs.portfoliointel.co.in/api-reference/upload-sessions) are owned by the key that created them, single-use, and expire after 15 minutes.

## 2026-09 — Enrichment

- Benchmark returns are measured as of the statement date, from index-fund NAVs.
- Scheme, category and benchmark matching use typed judgments, so a fund's category keeps its provenance.

## 2026-07-04 — MCP server fix

- `@portfoliointel/mcp-server` sends `tools/call` to `/mcp` as JSON-RPC. Update if tool calls failed from the stdio server.

## 2026-06-20 — Agent self-serve keys

- New unauthenticated [`POST /api/v1/register`](https://docs.portfoliointel.co.in/api-reference/register): an agent issues itself a `sk_test_` key with 100 credits and a claim link. Discoverable from `/.well-known/mcp/server-card.json`.

## 2026-06-02 — NSDL e-CAS

- NSDL e-CAS statements parse, including the CDSL demat accounts they list. See the [CAS format guide](https://docs.portfoliointel.co.in/reference/cas-format-guide).

## 2026-04-18 — Docs v2

- Docs migrated to Astro Starlight with deeper IA and search
- Tabbed code samples on quickstart and API reference
- 6-section sidebar: Get started / Platforms / API Reference / MCP / Skills / Reference

## 2026-04 — portfolio-health-check skill

- New skill endpoint: `POST /api/v1/skills/portfolio-health-check/run`
- 4 deterministic checks: concentration, AMC concentration, overlap, benchmark underperformance
- Exposed via MCP as `run_skill` tool

Source: https://docs.portfoliointel.co.in/changelog/
