MCP Tools
MCP tool reference, code samples, and example workflows.
Available MCP tools
Section titled “Available MCP tools”There are two servers: the hosted server at https://api.portfoliointel.co.in/mcp, and the @portfoliointel/mcp-server stdio package, which calls the same API. They expose slightly different tools. The stdio package is not yet published to npm, so connect to the hosted server, directly or through npx -y mcp-remote https://api.portfoliointel.co.in/mcp for clients that only run stdio servers.
| Tool | Hosted | npm | Description |
|---|---|---|---|
register | — | Yes | Get a free test key without signing up. Calls POST /api/v1/register. Hosted clients get a key from the OAuth consent page instead. |
parse_cas | Yes | Yes | Parse a CAS PDF and return structured holdings and metadata. Accepts pdf_url, pdf_base64 or session_id, plus pdf_path for a local file in the npm package. |
enrich | Yes | Yes | Add returns and benchmark overlays to holdings. |
request_upload_url | Yes | Yes | Create a browser-upload session when the agent can’t pass the file itself. |
list_skills | Yes | Yes | List available analysis skills. Needs no scope. |
run_skill | Yes | Yes | Run an analysis skill against a parsed snapshot. Returns flagged holdings with severity, reasoning and a not-advice disclaimer. |
generate_report | Yes | — | Export a snapshot as JSON or CSV. |
compare_snapshots | Not yet | — | Compare two snapshots for allocation drift and value change. Not listed by either server; a direct call returns 501 NOT_IMPLEMENTED. |
Prompts
Section titled “Prompts”Both servers also expose prompts: ready-made workflows that clients such as Claude Desktop show as slash commands. Clients that only support tools, including the Claude.ai web connector and ChatGPT, don’t show them.
| Prompt | Arguments | What it does |
|---|---|---|
analyze_my_portfolio | — | Gets the CAS (URL, local path, or upload link), parses it, enriches it, and summarises value, returns and top and bottom performers. |
portfolio_health_check | snapshot_id (optional) | Composition, returns, concentration and overlap, ending with the 2–3 findings with the largest effect. |
compare_to_benchmarks | snapshot_id (optional) | A 1Y and 3Y fund-versus-benchmark table, grouped into funds ahead on both horizons, behind on both, and mixed. |
tax_planning_india | snapshot_id, financial_year (both optional) | Indicative realised and unrealised STCG and LTCG by tax class, from transactions. Needs a detailed CAS. Not tax advice. |
Leave snapshot_id blank to have the prompt parse a new CAS first.
The prompts ask the model to state findings with numbers and not to tell the user to buy, sell, hold or switch a named scheme. Portfolio Intel is not a SEBI-registered investment adviser.
Getting the CAS to the agent
Section titled “Getting the CAS to the agent”If the user doesn’t have the PDF at hand, see Get your CAS. It explains how to request one and how an agent that already has the user’s own Gmail access can find the statement email and pass the attachment to parse_cas.
Skills
Section titled “Skills”Skills are higher-order analysis workflows that chain tools and apply domain-specific evaluation criteria. Unlike raw tools, skills produce structured reports with severity ratings and plain-English reasoning. See the Skills page for details, available skills, and usage examples.
Example workflow
Section titled “Example workflow”User asks an agent to parse a CAS and summarize the portfolio.
1. parse_cas → get snapshot_id and holdings2. enrich → attach market context3. agent summarizes current value, diversification, and available next actionsNode.js example
Section titled “Node.js example”const fs = require('fs');
async function parseCAS(filePath, password) { const pdfBase64 = fs.readFileSync(filePath).toString('base64');
const res = await fetch('https://api.portfoliointel.co.in/api/v1/parse', { method: 'POST', headers: { Authorization: 'Bearer sk_live_YOUR_KEY', 'Content-Type': 'application/json', }, body: JSON.stringify({ pdf_base64: pdfBase64, password }), });
const body = await res.json(); if (!body.success) throw new Error(body.error.message); return body.data;}Python example
Section titled “Python example”import base64import requests
def parse_cas(file_path: str, password: str | None = None) -> dict: with open(file_path, 'rb') as handle: pdf_base64 = base64.b64encode(handle.read()).decode()
resp = requests.post( 'https://api.portfoliointel.co.in/api/v1/parse', headers={'Authorization': 'Bearer sk_live_YOUR_KEY'}, json={'pdf_base64': pdf_base64, 'password': password}, ) resp.raise_for_status() return resp.json()['data']cURL sandbox example
Section titled “cURL sandbox example”# Sandbox samples are served by the web host, not api. — see Parse CAS → Sandbox samplescurl -X POST https://portfoliointel.co.in/api/v1/parse \ -H "X-API-Key: sk_test_YOUR_KEY" \ -H "Content-Type: application/json" \ -d '{"pdf_ref":"sample_cams_2024"}'