Use ConCallIQ’s memory layer with your AI tools.

Connect an MCP client to structured company memory, promise history, guidance, risks, and source references. ConCallIQ returns the record and its limits; the consuming model decides how to use it.

Hosted rollout status

The v2 interface is code-complete and preview-ready. Hosted production access is still gated while real-user auth verification, durable rate limits, and audit telemetry are completed.

Overview

What is ConCallIQ MCP?

ConCallIQ is a permanent memory layer for Indian company management. MCP makes that memory available to AI clients through bounded, deterministic tools instead of page scraping or free-form database access.

The architecture is intentionally transport-neutral: the same canonical query layer powers Ask ConCallIQ, local MCP, hosted MCP, and a future REST surface.

Memory

Query layer

MCP tools

Evidence

Tool surface

Available tools

Every tool is read-only. Basic retrieval is deterministic and does not invoke MiniMax or another model just to rephrase structured data.

Free hosted access

5 tools

search_companiesget_company_overviewget_available_quartersget_quarter (latest)get_evidence (latest)

Latest-quarter access keeps the first external connection useful while protecting deeper historical memory.

Pro / local access

8 tools

quarterly diffpromise ledgermanagement receiptsguidance and risksmanagement profilecompany comparisonpromise searchupcoming deadlines

The complete read-only surface remains source-first and never turns records into investment recommendations.

Resources include concalliq://company/{company}, quarter coverage, and Pro promise history. Prompt templates are available to Pro and local clients.

Provenance

Evidence and uncertainty

Significant records carry evidence IDs and canonical ConCallIQ/source URLs. IDs are validated server-side; callers cannot supply arbitrary URLs or filesystem/database identifiers.

Preserve assessment quality

For example, the current TCS record has 18 records carrying a missed assessment. Seventeen are heuristic rather than fully verified. A consuming model should say exactly that—not flatten the distinction into “TCS missed 18 promises.”

Human verified

Directly reviewed or supported by a verified source.

Heuristic

A structured assessment with an explicit quality flag.

Unspecified

Incomplete evidence; never silently converted to zero.

Setup

Connect a client

Local stdio is available for development. The hosted endpoint is documented here for preview configuration; do not put production credentials into client config files.

Local stdio

{
  "mcpServers": {
    "concalliq": {
      "command": "pnpm",
      "args": ["mcp:start"],
      "cwd": "/path/to/ConCallIQ.in"
    }
  }
}

Hosted preview

{
  "mcpServers": {
    "concalliq": {
      "url": "https://concalliq.in/api/mcp"
    }
  }
}

Hosted authentication uses a Supabase bearer token and canonical entitlement checks. The authorization-server configuration is intentionally not represented as complete until it is safely deployed.

Security

A constrained interface by design

No generic SQL, filesystem, or URL-fetch tools.

No mutation, billing, watchlist, email, or admin tools.

Origin, host, body-size, query, and output limits.

Stored evidence is treated as untrusted research content.

Free hosted history is limited to the latest quarter.

No Supabase service-role key is exposed to MCP.

Rollout

Preview first, public later

Before public hosted launch, ConCallIQ will validate real Free, Pro, expired, and invalid Supabase sessions; confirm RLS boundaries; add durable per-user rate limits; and record privacy-safe operational telemetry.

See the roadmap Read methodology