Login Schedule a Demo Request a Trial

Developers

Your agent can't quote a number without its source.

A native MCP server and bulk APIs that put cited filing intelligence inside your own agents and internal tools. The citation is part of the payload, not a footnote someone adds later.

Request a trial Contact us
  • 10MCP data servers
  • citeon every returned value
  • 1994–bulk backfill
  • read-onlyauto-approvable
  • RESTbulk and streaming
  • 0pipelines to maintain
The problem

An agent that can't cite is an agent you can't ship

Wire a model to a data feed and it will answer confidently with a number nobody can trace. The reviewer asks where it came from and the pipeline has no answer. The citation has to be in the payload, not bolted on after.

What you get instead

  • MCP servertools your agent can call directly
  • Provenance in the payloadcitations travel with the data
  • Bulk and streaming APIsbackfill and keep current
  • Read-only by designsafe to auto-approve

One call, end to end

the citation inside the payload, not beside it

  1. 01 · The call

    get_metric("NWR", "shares_outstanding")

    your agent calls a tool, not a scraper

    no pipeline to maintain

  2. 02 · The payload

    { "value": 1268500, "cite": { "filing": "10-Q", "para": 18 } }

    as_of travels too — the source is a field, not a footnote

  3. 03 · The filing

    10-Q · paragraph 18 · as filed

    Accession 0001234567-24-000091

    Same reference in bulk and streaming

    identical citation across every surface

  4. 04 · The answer

    The model quotes with a reference attached

    call cite answer

    reviewable without re-deriving it

What's in the solution

Four jobs this covers end to end.

01 · Agents

Tools your model can call

A native MCP server that works with Claude, ChatGPT, Gemini and Cursor — connect it and the tools appear.

MCP

02 · Provenance

The cite is a field

Every returned value carries the filing and paragraph it came from, so anything the model says can be checked without re-deriving it.

Provenance

03 · Data

Bulk and streaming

Backfill the history, then stay current — the same schema and the same citation reference across both.

APIs

04 · Safety

Read-only by design

Nothing writes, nothing mutates. Safe to auto-approve in an agent loop without a review step on every call.

Read-only

How it lands on your desk

Same data, three surfaces — pick by where the work happens.

AI engineers

The scenario

You are building an internal agent and the blocker is not the model — it is giving it data a reviewer will accept.

MCP ServerAPIsAI Chat

How Kaleidoscope helps

  • Tools appear the moment the server is connected
  • Citation in the response schema, not a post-processing step
  • Read-only tools you can auto-approve in a loop
  • Connect only the domains you licence
  • No retrieval pipeline to build or keep fresh
The products

Three surfaces onto the same data — pick the one that fits your stack.

MCP Server

The agent surface

Connect any subset of ten domain servers to your AI client. Your agent queries filings, deals, ownership and the rest directly.

  • Works with Claude, ChatGPT, Gemini and Cursor
  • Connect only the domains you licence
  • Screen filings and run signal queries without a pipeline
  • Read-only, auto-approvable, cited on every response
Explore MCP Server → 10data servers
mcp · kscope

Agent calls

insider_history(ticker="CRWD", limit=1)

MCP returns

{ "issuer": "CrowdStrike Holdings, Inc.",
  "role": "Chief Financial Officer",
  "date": "2026-09-24", "type": "S",
  "shares": -18000, "price": 263.51,
  "plan_10b5_1": true,
  "source": { "form": "4" } }

The source travels with the value — the agent can't quote the trade without its filing.

An MCP tool call — the Form 4 comes back with its source attached

AI Chat

The hosted surface

The same tools in a hosted chat, for teams who want cited answers today without wiring anything up first.

  • The same tools as the MCP server, already connected
  • Answers carry the filing reference
  • No client configuration to manage
  • A way to try the tools before you build on them
Explore AI Chat → citeon every answer
Argos, the hosted research agent — every claim backed by a sources panel

Argos, the hosted research agent — every claim backed by a sources panel

APIs

The data surface

REST and bulk endpoints for teams that want the data in their own warehouse rather than in an agent loop.

  • Bulk history for backfill, streaming to stay current
  • Same schema and citation reference as MCP
  • Stable identifiers across every domain
  • Point-in-time values, never restated
Explore APIs → 1994–bulk backfill

GET

/v1/funding/rounds?company=black-forest-labs

Response

[ { "round": "Series B", "amount_usd": 300000000,
    "announced": "2025-12-01", "source": "press release" },
  { "form": "D", "filed": "2026-06-03",
    "exemption": "506(b)", "offering_usd": 343309894,
    "sold_usd": 307572022, "source": "SEC EDGAR" } ]

Press and filing on one company ID — every row says where it came from.

The same record over REST — press round and Form D on one company

Every product in this solution

Start with the MCP server; move to bulk when the warehouse becomes the target.

Full data catalog →

Core platform

  • MCP Server Ten domain servers your agent can call Integration
  • APIs Bulk backfill and streaming, one schema REST · Bulk
  • AI Chat The same tools in a hosted chat MCP · Chat
  • SEC Pro+ The same data with a UI on it Research UI

MCP data layers · connect any subset

  • SEC Filings, 13F flow, activist campaigns, ADV, XBRL Live
  • SEDAR Canadian mining from NI 43-101 technical reports Live
  • Funding Private rounds, investors, valuations Live
  • Mergers Announced and closed deals, terms, comparables Live
  • Agreements Material contracts, cited to the clause Live
  • News 9-class taxonomy, materiality, price correlation Live
  • Catalyst Earnings, scheduled events, macro releases Live
  • Bio Point-in-time drug pipelines and base rates Live
  • Power Generation, capacity, the AI power-demand trade Live
  • Tranche Issuer debt, bond cards, refinancing screens Live

Workflow & delivery

  • Reports Generated output with citations attached HTML · MD · JSON
  • Alerts Push triggers into your own systems Webhooks
  • NexusID Stable identifiers across every domain Resolution
  • Provenance Core The classifier engine behind the signals Signals

Connect it to your AI client

Claude · ChatGPT · Gemini · Cursor
  • screen_filings
  • get_metric
  • run_signal_query
  • graph_resolve
  • get_filing_history
  • read-only · auto-approvable
What did this issuer say about termination rights last year versus this year, and cite both.

→ returns both passages with accession and paragraph references.

Or take it as data

  • Bulk history back to 2014, point-in-time and never restated
  • Streaming updates on the same schema
  • Citation reference carried on every record
How it compares Kaleidoscope Generic data APIs Build it yourself
Citations A field in every response Not modelled Whatever you capture
Agent access Native MCP server Wrap it yourself Wrap it yourself
Coverage Ten domains on one schema Usually one domain As many as you build
History Point-in-time, never restated Often revised silently Yours to guarantee
Approval Read-only, auto-approvable Depends on the wrapper Yours to design
Maintenance No pipeline to keep fresh Client code only A standing cost

Why teams switch

We did not need a better model. We needed the number to arrive with a filing reference so the reviewer would sign off.
the cite is a fieldread-only, auto-approvableten domains, one schema

From the research desk

Ask Anything: SEC research, finally conversational with our MCP server The diligence gap: four questions a strong LLM still gets wrong

Build with cited data

Talk to our team about how you work today, or start a trial and explore the platform yourself.

Contact us Request a trial

Related solutions

Hedge Funds, Quants & Data Teams → VC, Family Offices & PE → Legal & Compliance →