Funnel Calibrator
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OBSIDIAN_PORT | No | Port for the Obsidian Local REST API (default 27123). | 27123 |
| OBSIDIAN_API_KEY | No | API key for the Obsidian Local REST API plugin (needed for the agent flow). | |
| KEYCRM_MCP_SECRET | No | Shared secret for accessing the KeyCRM MCP worker when regenerating the dataset. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| measure_sku_funnelA | Measure the observed call-centre approval, post-office buyout, and return rates for one product over a date window, counting only order cohorts old enough to have resolved. Use this before any decision that depends on approval or buyout rates, instead of assuming the portfolio-wide 65% approval and 52.5% buyout. Returns rates together with sample size and a reliability flag, and returns null rates with reliability 'insufficient' rather than a number when the resolved sample is too small to support a conclusion. Read-only. |
| recalibrate_cpl_boundsA | Recompute Stop CPL (the cost per lead at which profit reaches zero) and Goal CPL (the optimisation target) for one product from its own measured funnel rates instead of portfolio-wide assumptions, and report the drift against the baseline currently in use. Measures the product itself when observed rates are not supplied. Use whenever a cost-per-lead target is being set, defended, or questioned. Flags the case where the assumed target sits above the true break-even, which is invisible in the ads dashboard. Read-only. |
| recommend_next_actionA | Diagnose why a product is underperforming and return the matching remedy. Given the product's measured funnel, its current cost per lead, and optional competitor and creative signals, identifies the likeliest failure mode — structural loss, contested auction, traffic quality, offer or price mismatch, creative fatigue, or a weak offer — and returns the action that fits it. Use when a campaign misses its target and the cause is not yet established, because cheap leads that fail on the call and leads that fail at the post office need opposite remedies. Returns action 'hold' with diagnosis 'insufficient_data' rather than guessing when the sample is too small. Read-only. |
| audit_ad_verdictA | Test an advertising decision that originated elsewhere — a daily watchdog report, an operator's judgement, a note in the decision journal — against this product's recalibrated economics. Returns 'supported', 'contradicted', or 'insufficient_data', together with the numeric chain that produced the verdict and a counter-recommendation where the proposal is contradicted. Use before acting on any recommendation this server did not itself produce. Read-only. |
| list_covered_skusA | List the product codes present in the loaded snapshot, with their order counts and whether price and cost data exist for them. Use to resolve a product code before measuring, or to see what the dataset covers. Read-only. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 5 tools
Each tool targets a distinct decision stage: listing covered SKUs, measuring observed funnel rates, recalibrating CPL bounds, recommending remedies, and auditing external verdicts. The two diagnostic tools (recommend_next_action and audit_ad_verdict) are clearly separated by whether the decision originated from this server or elsewhere. No overlapping or indistinguishable purposes.
All tools use a consistent snake_case verb_noun pattern (measure_sku_funnel, recalibrate_cpl_bounds, recommend_next_action, audit_ad_verdict, list_covered_skus). The convention is predictable and readable throughout.
Five tools is well-scoped for a specialized funnel calibration server. Each tool earns its place: discovery, measurement, recalibration, diagnosis, and verification. There is no redundancy or thin surface.
The surface covers the full read-only analytics lifecycle from listing SKUs to auditing external decisions, with sensible fallbacks for insufficient data. A minor gap is the absence of a dedicated tool to fetch or compare the current portfolio-wide baseline assumptions, though descriptions imply those are internally available.