Skip to main content
AI agents pay for tokens. SEC API is designed to minimize token consumption while maximizing information density.

​
The Token Tax

When an AI agent queries SEC data, every byte of the API response consumes tokens. Bloated responses waste money and slow down agent reasoning. SEC API solves this with compact, purpose-built responses.

​
Headline numbers

Two effects compound:
  • vs raw SEC EDGAR documents: roughly 21x fewer tokens. Pulling a filing from EDGAR and feeding the raw HTML/XBRL to an agent costs an order of magnitude more tokens than the structured, pre-extracted response SEC API returns for the same question.
  • ?view=agent trims about 93% of the response envelope versus the default view on envelope-heavy endpoints (for example, entities.resolve drops from ~4.9 KB to ~0.3 KB), while preserving the citation fields agents need.
These are workload-specific measurements, not universal guarantees — re-run them with the Measure It Yourself snippet below against your own queries.

​
Side-by-Side: SEC API vs sec-api.io

Competitor numbers below were captured 2026-03-18; re-capture before citing as current.

​
Entity Resolution

MetricSEC APIsec-api.ioSavings
Response size273 bytes412 bytes34%
Estimated tokens6810334%
Latency62ms231ms73%
MetricSEC APIsec-api.ioSavings
Response size500 bytes792 bytes37%
Estimated tokens12519837%
Latency64ms281ms77%

​
Section Extraction

MetricSEC APIsec-api.ioSavings
Response size1,800 bytes2,880 bytes38%
Estimated tokens45072038%
Latency64ms348ms82%

​
Intelligence Bundles: 75% Token Reduction

A typical “company briefing” requires assembling data from multiple sources:

​
Traditional Approach (sec-api.io)

1. Entity resolve         →  103 tokens
2. Latest 10-K search     →  198 tokens
3. Item 1A extraction     →  720 tokens
4. Latest 10-Q search     →  198 tokens
5. Executive compensation →  450 tokens
6. Insider trades          →  350 tokens
7. 13F holdings           →  400 tokens
8. Recent 8-K events      →  600 tokens
─────────────────────────────────────
Total: 8 API calls, ~3,019 tokens

​
SEC API Intelligence Bundle

1. Company intelligence   →  ~800 tokens
─────────────────────────────────────
Total: 1 API call, ~800 tokens
Result: 75% fewer tokens, 87% fewer API calls.

​
Why SEC API Responses Are Smaller

  1. No wrapper bloat. Responses contain data, not framework metadata.
  2. Agent mode. Use ?view=agent to get token-efficient responses that preserve citation fields where applicable.
  3. Pre-computed intelligence. One bundle call replaces 8+ raw API calls.
  4. Semantic search. Find relevant content in one call instead of paginating through keyword results.

​
Field selection and tabular output (MCP)

Over MCP, tools/call accepts two params-level controls (siblings of name and arguments, not inside arguments). Both shape only the human/agent-facing text content — structuredContent always stays canonical JSON, so typed clients and outputSchema validation are unaffected.

​
params.fields — keep only the fields you asked for

Pass a comma-separated list of dotted paths to project the text content down to those fields. For a list result the projection is applied to each row in data; envelope keys (object, requestId, traceparent, and the list keys data/hasMore/nextCursor/degradedState) are always retained so the result stays schema-valid. Paths address object fields — the projection does not descend into nested arrays. 
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "companies.overview",
    "arguments": { "tickers": ["AAPL", "MSFT", "NVDA"] },
    "fields": "identity.ticker,financialSnapshot.metrics.revenue.value"
  }
}

​
params.response_format — compact columnar or CSV for lists

json (default) returns the full object form. table returns a compact columnar envelope ({ object: "table", columns, rows }) that lists the columns once and emits one array of cells per row, dropping the repeated per-row keys; csv returns RFC-4180 CSV text. Both apply to the data array of list-shaped results and shine on flat rows; non-list results fall back to JSON. Decoding is trivial — pair each columns[i] with rows[n][i]. 
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "insiders.list",
    "arguments": { "ticker": "META" },
    "response_format": "table"
  }
}
fields and response_format compose: projection runs first, so a table’s columns are exactly the (flat) fields you kept. This pairs naturally with the multi-entity batch tools (companies.overview, companies.financials, companies.ratios accept a tickers array of up to 50) — request a portfolio in one call, then trim and tabularize it.

​
include= — REST opt-in for expensive sub-objects

include= is a REST query param, not an MCP control: where fields subtracts, include adds bounded enrichments that are omitted by default — for example GET /v1/companies/overview?include=segments,footnotes,dilution,factors. Omitted enrichments report status: not_requested rather than being computed.

​
Measure It Yourself

The ?view=agent reduction is the easiest to reproduce — compare the agent view against the default view of the same endpoint: 
# view=agent vs default response size (same endpoint, same key)
AGENT=$(curl -s "https://api.secapi.ai/v1/entities/resolve?ticker=AAPL&view=agent" \
  -H "x-api-key: $SECAPI_API_KEY" | wc -c)

DEFAULT=$(curl -s "https://api.secapi.ai/v1/entities/resolve?ticker=AAPL" \
  -H "x-api-key: $SECAPI_API_KEY" | wc -c)

echo "view=agent: $AGENT bytes, default: $DEFAULT bytes"
# Example result: view=agent ~339 bytes vs default ~4880 bytes (~93% smaller)
To compare against sec-api.io, use your own sec-api.io key (their endpoints require authentication and reject anonymous requests): 
SEC_API_SIZE=$(curl -s "https://api.sec-api.io/mapping/ticker/AAPL?token=$SEC_API_KEY" | wc -c)
echo "sec-api.io: $SEC_API_SIZE bytes"
For exact output-token telemetry, add SECAPI-Compute-Headers: token-count to the request. Every response includes a SECAPI-Estimated-Cost header; cache hits also return an exact SECAPI-Token-Count without recomputing the body.