agent-reasoning-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| set_goalA | Register, update, decompose, inspect, or abandon hierarchical goals and task DAGs (actions: create, update, decompose, get, list, abandon). Use set_goal instead of manage_intentions when formulating multi-step objectives and sub-goal DAGs rather than queueing concrete execution directives. Returns goal record, sub-goal hierarchy, completion progress, or goal lists. |
| evaluate_situationA | Ingest multi-modal situation snapshot, compute expected utilities against active weights, and output prioritized action recommendations (actions: snapshot, quick). Use evaluate_situation instead of assess_risk when ranking candidate actions across multi-attribute utility dimensions rather than calculating isolated threat probabilities. Returns ranked candidate actions, expected utility scores, and top recommendation. |
| replanA | Regenerate sub-task DAG and adjust intentions upon unexpected obstacles or state changes (actions: blocker, event, full). Use replan instead of set_goal when recovering from execution blockers or environment shifts rather than creating new goals. Returns replanned goal DAG, invalidated intentions, and newly synthesized subgoals. |
| assess_riskA | Compute quantitative risk and threat assessment for candidate actions, plans, or 3D spatial rollouts against active utility weights (actions: action, plan, compare, spatial_rollout). Use assess_risk instead of evaluate_situation when estimating failure probability and threat exposure rather than ranking overall utility. Returns risk score (0.0-1.0), threat breakdown, and comparative risk ratings. |
| query_knowledgeA | Search learned heuristic patterns, tactics, and past decision traces by context similarity (actions: search, patterns, similar_situations). Use query_knowledge instead of get_decision_trace when retrieving generalized patterns across sessions rather than inspecting a single execution trace. Returns matching heuristics, anti-patterns, tactics, and similarity scores. |
| set_utility_weightsA | Configure, inspect, activate, or incrementally nudge multi-attribute utility weight profiles (actions: configure, get, list, activate, nudge). Use set_utility_weights instead of evaluate_situation when defining decision preferences (aggression, caution, greed, exploration) rather than evaluating actions. Returns configured utility profile, active weight map, or profile directory. |
| get_decision_traceA | Retrieve explainable step-by-step chain-of-thought rationale, candidate utilities, and risk assessment for past decisions (actions: latest, get, list, explain). Use get_decision_trace instead of query_knowledge when performing deep forensic analysis of a specific historical decision. Returns step-by-step reasoning trace, utility breakdown, candidate rankings, and explanation text. |
| manage_beliefsA | Maintain structured belief state with TTL expiration sweeps, exponential confidence decay, and category filtering (actions: update, query, expire, reconcile, reconcile_spatial). Use manage_beliefs instead of query_knowledge when managing dynamic agent epistemic state rather than static heuristic patterns. Returns belief record, query matches, expired belief count, or reconciliation report. |
| manage_intentionsA | Queue, dispatch, track, cancel, or resolve behavior directives for behavior-mcp runtime execution (actions: create, dispatch, get, list, cancel, resolve). Use manage_intentions instead of set_goal when dispatching immediate execution instructions to runtime behaviors rather than managing abstract objectives. Returns intention record, dispatch status, wire contract payload, or intention list. |
| manage_reasoning_dbA | Database maintenance, diagnostics, SHA-256 Merkle audit verification, and snapshot management (actions: stats, audit, doctor, snapshot, diff, restore). Use manage_reasoning_db instead of manage_beliefs when performing SQLite storage integrity verification or database snapshot restore. Returns database diagnostics, Merkle audit tree, snapshot metadata, or diff reports. |
| classifyA | Assign semantic categorical labels to an entity, visual state, task, or state snapshot using deterministic System One calculus. Use classify instead of ask_choice when assigning predefined taxonomy labels rather than selecting among runtime decision alternatives. Returns top class label, probability distribution, and classification margin. |
| ask_noulA | Evaluate whether a specific proposition is true given the current state pack with calibrated probability and abstain safeguards. Use ask_noul instead of ask_score when evaluating binary truth/falsehood rather than scoring an entity on a continuous scale. Returns boolean answer, confidence probability, and abstain flag. |
| ask_choiceA | Select 1 option from a discrete set of alternatives (N <= 16) with probability distribution and utility margin. Use ask_choice instead of classify when choosing the best action or alternative under active utility profiles rather than categorizing an entity. Returns selected option ID, probability distribution, and utility margin. |
| ask_scoreA | Evaluate an entity, plan, or action on a bounded continuous scale against weighted criteria. Use ask_score instead of ask_noul when evaluating continuous numeric quality or fitness rather than binary truth. Returns normalized score within scale bounds, criterion breakdown, and evaluation confidence. |
| gate_intentionA | Evaluate an intention before dispatching to behavior-mcp and issue a cryptographic HMAC dispatch token if approved. Use gate_intention instead of manage_intentions when verifying precondition safety and issuing execution authorization rather than tracking intention state. Returns gate verdict (approved/rejected), risk evaluation, and HMAC dispatch token. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| strategic-assessment | Generate comprehensive strategic assessment from active goals, beliefs, and situation snapshot |
| goal-planning | Decompose high-level strategic objectives into structured sub-goal DAGs with verifiable success criteria |
| risk-evaluation | Perform quantitative threat and opportunity analysis for proposed action plans |
| post-mortem | Analyze decision trace and outcome results to extract heuristic knowledge patterns and lessons learned |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| reasoning-health | Server health status, version, and timestamp |
| docs-set_goal | Complete parameter schema and documentation for set_goal |
| docs-evaluate_situation | Complete parameter schema and documentation for evaluate_situation |
| docs-replan | Complete parameter schema and documentation for replan |
| docs-assess_risk | Complete parameter schema and documentation for assess_risk |
| docs-query_knowledge | Complete parameter schema and documentation for query_knowledge |
| docs-set_utility_weights | Complete parameter schema and documentation for set_utility_weights |
| docs-get_decision_trace | Complete parameter schema and documentation for get_decision_trace |
| docs-manage_beliefs | Complete parameter schema and documentation for manage_beliefs |
| docs-manage_intentions | Complete parameter schema and documentation for manage_intentions |
| docs-manage_reasoning_db | Complete parameter schema and documentation for manage_reasoning_db |
| docs-classify | Complete parameter schema and documentation for classify |
| docs-ask_noul | Complete parameter schema and documentation for ask_noul |
| docs-ask_choice | Complete parameter schema and documentation for ask_choice |
| docs-ask_score | Complete parameter schema and documentation for ask_score |
| docs-gate_intention | Complete parameter schema and documentation for gate_intention |
TDQS
Scored across 15 tools
Most tools target distinct concepts, and the descriptions actively disambiguate with explicit 'use X instead of Y' guidance (e.g. assess_risk vs evaluate_situation, set_goal vs manage_intentions). However there is a dense cluster of risk/utility evaluation tools (assess_risk, evaluate_situation, ask_score, gate_intention) that an agent could still misselect among.
The set consistently uses snake_case verb_noun patterns (manage_beliefs, query_knowledge, get_decision_trace, set_utility_weights), with coherent families like ask_* and manage_*. Minor deviations are bare verbs (replan, classify) and an unusual token (ask_noul), but overall the convention is predictable.
15 tools sit at the upper edge of the comfortable 3-15 range but are justified by a genuinely broad cognitive-architecture domain (risk, beliefs, goals, intentions, knowledge, DB ops). It is slightly heavy but each tool maps to a real capability rather than being filler.
Coverage is strong: beliefs (update/query/expire/reconcile), goals (create/update/decompose/get/list/abandon), intentions (create/dispatch/get/list/cancel/resolve) and traces/knowledge all form coherent lifecycles. Gaps are minor—no obvious delete or bulk-export for some resources—but core reasoning workflows are well covered.