Skip to main content
Use the hosted MCP server when an agent client needs to discover and call SEC API tools. SEC API provides the MCP transport and the deployed tool inventory; it does not certify an agent’s summary. Use REST or an SDK when your server owns the workflow directly or needs a documented HTTP operation that is not advertised by MCP discovery.

Prove one agent task can return a source

Connect one supported MCP client using an environment-backed API key. Discover the available tools, resolve an issuer, retrieve a filing and section, then cross-check the same record over REST. A useful result is a reproducible filing record, not just a fluent response.

Configure the client and verify the record

Discovery also requires an API key. Send it from a server-side secret or MCP client configuration, not from a browser bundle:
Configure the client to send x-api-key from its secret environment. Ask it to discover tools, then use the deployed equivalents of issuer resolution, filing retrieval, and section retrieval for AAPL and Item 1A. The authenticated discovery response for that session is the tool contract to test. Cross-check the selected source over REST:
GET /mcp returns discovery metadata only when the request is authenticated. JSON-RPC tool calls use POST /mcp with the same secret-backed API key.

Evidence standard

Retain a redacted client configuration record, client name, discovered tools, arguments, accession number, filing URL, section key, returned text, requestId, traceparent, and available provenance or freshness. The REST and MCP records should identify the same source. They need not produce identical agent wording.

Choose the right integration boundary

MCP is useful when a client can preserve source identity and surface a failed tool call. Keep the credential and retrieval loop behind your own server when the client cannot do that, or when you need direct control over retries, request shaping, or the full REST surface. Filing text is source material, not legal, accounting, or investment advice; use a human-review path when your application requires one.

Next step

Enable one bounded agent task only after the source record is visible to a reviewer. Treat authentication, authorization, validation, quota, and execution failures as separate outcomes before broadening the tool set.

Install MCP

Set up an environment-backed client.

MCP workflows

Review discovery and evidence patterns.

API conventions

Use REST when you own the integration.

Create an API key

Use a secret-backed credential.

Open the sandbox

Inspect the REST evidence path.

Public pricing

Review agent-loop usage.
1

Read discovery

Configure the client from authenticated discovery metadata.
2

Call a small tool chain

Require filing and section identity.
3

Cross-check source

Confirm the source over REST.