Can the MCP server change anything?

No. Every tool is read-only. It cannot create or edit budgets, keys, routing or any other setting.

Which plan includes it?

Base, Growth and Business (and the 10-day trial). The compare_variants tool needs Growth.

Is this the same as the MCP Gateway?

No. The MCP Gateway (Business) sits in front of your own MCP servers and governs your agents' tool calls. The MCP server described here exposes Cognocient's own data to your assistant.

Which key should I use?

A Cognocient proxy key. Create one per person (Settings → Proxy Keys) so each can be revoked on its own.

How do I ask about my AI spend from Claude Code or Cursor?

Cognocient's read-only MCP server lets Claude Code, Cursor, Claude Desktop or any MCP client query your spend, budgets, variant comparisons and CI cost-gate results. Base plan.

In the app: Integrations → MCP server · Base plan and aboveOpen in app →No account? Start free

Ask your coding assistant questions like "what did the checkout feature spend in production last week?" or "did the last CI run cost more than main?" and get answers from your real Cognocient data.

Connect

The server speaks MCP over Streamable HTTP at:

https://api.cognocient.com/api/mcp

Claude Code

claude mcp add --transport http cognocient https://api.cognocient.com/api/mcp \
  --header "Authorization: Bearer <your Cognocient proxy key>"

Cursor, Claude Desktop and other clients (JSON config)

{
  "mcpServers": {
    "cognocient": {
      "url": "https://api.cognocient.com/api/mcp",
      "headers": { "Authorization": "Bearer <your Cognocient proxy key>" }
    }
  }
}

Tools

Tool	What it returns
`get_spend_summary`	Total spend and calls for a period, top 5 models and features
`get_spend_breakdown`	A ranking by feature, model, provider, department, environment, variant, user, project or workload, for spend, calls, tokens, average cost per call, error rate or p95 latency
`get_spend_trend`	A daily (or hourly, up to 7 days) series, optionally split by one dimension (top 5)
`list_budgets`	Every budget with its scope, limit, month-to-date spend, percent used and mode
`compare_variants`	The Experiments comparison by variant, model or environment (Growth)
`list_cost_gate_checks`	Recent CI cost gate results with reasons

Breakdowns and trends accept exact-match filters on any of the dimensions, for example {"environment": "production", "feature": "checkout"}. Use (untagged) to match calls without a tag. Periods run up to 90 days (7 on Free).

The call log holds usage and cost, not prompt or response text, so the assistant can answer "how much" and "where", not "what did the model say".

Details

Stateless: no sessions and no server-sent events. GET returns 405, which MCP clients handle.
Protocol versions 2024-11-05 through 2025-11-25 are accepted.
JSON-RPC batches are rejected (removed from MCP in 2025-06-18).
Errors in tool arguments come back as tool results with isError: true, so the assistant can correct itself.

How do I ask about my AI spend from Claude Code or Cursor?

Connect

Tools

Details

On this page