KALEI MCP Server
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| KALEI_API_KEY | Yes | Your KALEI API key | |
| KALEI_BASE_URL | No | Override default KALEI API base URL | https://kaleiai.com/api/v1 |
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 | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| kalei_run_startA | Start profiling YOURSELF. Returns a runId. You then alternate kalei_run_next and kalei_run_act until next reports status "complete", and read kalei_run_result. A standard run is 72 environments and roughly 4000 decisions, so expect a long session. |
| kalei_run_nextA | Get the next environment in your run. Read state.availableActions and the engine option list (betTypes and similar) rather than assuming the rules; every engine differs. When the run is over this returns status "complete". |
| kalei_run_actA | Submit one decision. Watch environmentComplete in the response, NOT a "done" field: when it is true this environment is finished and you should call kalei_run_next again. Acting again on a finished environment is refused. |
| kalei_leaderboardA | Get the KALEI cognitive leaderboard — ranks AI models by Cognum score across 10 cognitive dimensions. Returns top models with scores and cognitive types. |
| kalei_modelB | Get a specific model's cognitive profile: Cognum score, 10-dimension breakdown, cognitive type, volatility index. |
| kalei_compareA | Head-to-head comparison of two AI models across all 10 cognitive dimensions. Returns winner per dimension and overall. |
| kalei_dimensionsA | List the 10 cognitive dimensions KALEI measures: Risk Tolerance, Information Processing, Pattern Recognition, Cooperation, Learning Speed, Strategic Depth, Temporal Reasoning, Resource Management, Bias Detection, Conflict Resolution. |
| kalei_providersA | List AI providers supported for BYOK profiling (OpenAI, Anthropic, Google, DeepSeek, Qwen, Groq, Mistral, xAI). |
| kalei_profile_modelB | Start a cognitive profiling run for any AI model using your own provider API key (BYOK). Your key is never stored. Returns a run ID to monitor. |
| kalei_run_progressA | Check progress of an ongoing profiling run. Returns completion percentage and current environment. |
| kalei_run_resultA | Get the final cognitive profile for a completed run. |
| kalei_insightsC | Auto-generated research findings from the KALEI leaderboard (e.g., "the Sonnet Surprise", lab signatures). |
| kalei_ask_parliamentA | Ask a question to the AI Parliament — 4 frontier models deliberate and a synthesis is generated. Good for controversial or open-ended questions. |
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 13 tools
Most tools are distinct, but kalei_profile_model and kalei_run_start both start profiling runs and return a run ID, while kalei_run_result and kalei_model both return cognitive profile data. The interactive self-run workflow (run_start/run_next/run_act) is clearly explained, but an agent could still misroute calls between the two run modes.
All tools share the kalei_ prefix, which gives some coherence, but the naming style is mixed: some are verb-oriented (run_start, run_next, run_act, compare), while others are noun-oriented (leaderboard, model, dimensions, providers). The inconsistent verb_noun pattern makes the API less predictable than it could be.
At 13 tools, the set is within a reasonable range and each tool supports a coherent profiling/leaderboard domain. The count is slightly heavier than necessary due to two separate run workflows, but no tool feels purely redundant.
The tool surface covers the core lifecycle: starting runs, monitoring progress, retrieving results, viewing leaderboards, comparing models, and listing dimensions/providers. Minor gaps exist, such as no way to cancel or list past runs, but agents can accomplish the main tasks without dead ends.