Tools by Tier¶

G6Solver organizes its 262 MCP tools into 15 categories. Your license plan determines which categories you can access.

Plans¶

Plan	Categories	Tools	Rate Limit	Price
Free Trial	Core, Guide	30	10 req/min	30 days free
Researcher	+ Navigator, EvoSkill (read-only), Bug Reporting, Tier Status, Job Workbench (flagship workflows), Grounding	66	100 req/min	$20 AUD/mo
Builder	All 15 categories	Full tool access	500 req/min	$50 AUD/mo

Free Trial

Free Trial accounts receive the Researcher tool set for 30 days, at the Free Trial rate limit (10 req/min). After 30 days the account drops to the Free tool set.

Trial accounts do not receive Builder tools. They used to — an unpaid account held the full Builder surface, and held it indefinitely because nothing expired the trial. That was corrected on 2026-08-16 so that no unpaid account holds Builder access.

Category 1 -- Core Tools (9)¶

Available on all plans (Free Trial, Researcher, Builder). Fundamental operations for goal decomposition, ML, and system management.

Tool	Description	Component
`decompose_goal`	Break goals into executable subtasks	`goal_engine`
`run_bayesian`	Bayesian inference and posterior analysis	`adapt_bayesian`
`run_automl`	Automated machine learning pipeline search	`adapt_automl`
`recommend_algo`	Meta-learning algorithm recommendation	`adapt_sklearn`
`system_status`	System health and operational metrics	`core`
`list_components`	List all registered components	`core`
`invoke_component`	Call any component by name and operation	`core`
`invoke_component_typed`	Call any component with typed input validation	`core`
`run_pipeline`	Execute multi-step component pipelines	`core`

Example: decompose_goal

{
  "goal": "Build a fraud detection classifier",
  "max_depth": 3
}

Returns a search tree with subtasks, node count, and execution plan.

Category 2 -- Guide Tools (5)¶

Available on all plans (Free Trial, Researcher, Builder). Natural-language interaction and workflow planning for onboarding and discovery.

Tool	Description	Component
`guide_ask`	Ask questions about the system in natural language	`guide`
`guide_how_to`	Get step-by-step how-to guides	`guide`
`guide_find_tool`	Find the right tool for a given task	`guide`
`guide_plan_workflow`	Plan multi-tool workflows from a goal	`guide`
`guide_system_overview`	Get a high-level system overview	`guide`

Example: guide_plan_workflow

{
  "query": "Analyze CSV data, train a model, and verify safety",
  "top_k": 5
}

Returns pipeline steps, a Mermaid diagram, and component recommendations.

Category 3 -- Navigator Tools (6)¶

Available on Researcher and above. Discovery, search, and inspection of the component ecosystem.

Tool	Description	Component
`nav_discover`	Discover available capabilities by query	`navigator`
`nav_search`	Search components by capability keyword	`navigator`
`nav_recommend`	Get ranked component recommendations for a task	`navigator`
`nav_inspect`	Inspect a component's MCP tools and metadata	`navigator`
`nav_stats`	System-wide statistics (component counts, clusters)	`navigator`
`nav_health`	Health check across all registered components	`navigator`

Example: nav_recommend

{
  "query": "sentiment analysis on customer reviews",
  "limit": 5
}

Returns ranked component suggestions with relevance scores.

Category 4 -- Basic Extras (5)¶

Available on Researcher and above. Read-only access to EvoSkill, bug reporting, and tier status.

Tool	Description	Component
`list_programs`	List all evolved programs in the EvoSkill library	`evoskill`
`get_frontier`	Get the current Pareto frontier of evolved programs	`evoskill`
`patch_report_bug`	Report a bug against any component	`patch_client`
`patch_check_status`	Check the status of a reported bug	`patch_client`
`tier_get_status`	Get your current tier advancement status	`tier_manager`

Category 5 -- Template Pipelines (5)¶

Available on Builder only. Pre-built end-to-end pipelines for common workflows.

Tool	Description	Pipeline Steps
`run_research_pipeline`	End-to-end research pipeline	search -> RAG -> summarise
`run_codegen_pipeline`	Code generation pipeline	meta-program -> formal verify -> evaluate
`run_data_pipeline`	Data analysis pipeline	pandas -> sklearn -> evaluate
`run_safety_pipeline`	Safety verification pipeline	CSF check -> formal verify -> ground
`run_knowledge_pipeline`	Knowledge extraction pipeline	extract -> ground -> store

Example: run_research_pipeline

{
  "query": "Recent advances in transformer architectures"
}

Chains ctx_search, ctx_rag, and ctx_recursive in sequence.

Category 6 -- Orchestration (4)¶

Available on Builder only. Advanced composition, cross-component invocation, and introspection.

Tool	Description
`compose_and_run`	Compose and execute custom multi-step pipelines
`cross_call_tool`	Cross-component tool invocation with effect budgets
`dependency_graph`	View the full component dependency graph
`explain_component`	Get detailed explanations of any component

Category 7 -- EvoSkill (7)¶

Available on Builder only. Self-improving evolutionary skill synthesis and program evolution.

Tool	Description	Component
`evolve_skills`	Run evolutionary optimization on a skill population	`evoskill`
`propose_skill`	Propose a new skill for evolutionary evaluation	`evoskill`
`generate_skill`	Generate a skill implementation from a specification	`evoskill`
`evaluate_program`	Evaluate a program against fitness objectives	`evoskill`
`list_programs`	List all evolved programs in the library	`evoskill`
`get_frontier`	Get the current Pareto frontier of evolved programs	`evoskill`
`feedback_descent_run`	Run feedback-descent optimization on a skill	`evoskill`

Example: evolve_skills

{
  "skill_name": "sort_algorithm",
  "population_size": 20,
  "generations": 10,
  "objectives": ["speed", "correctness"]
}

Runs evolutionary optimization on a skill population, returning the Pareto-optimal programs.

Category 8 -- Self-Optimisation (4)¶

Available on Builder only. Bounded optimisation loop for pipeline speed, cost, and quality.

Production caveat

These tools can analyse workflow traces, recommend Pareto-valid interventions, apply selected plans through the safety gate, and record outcome data. Treat them as optimisation assistance with validation evidence, not as a guarantee that a production workflow will autonomously improve without review. Customer-facing claims should be backed by before/after measurements from the workflow being discussed.

Tool	Description	Delegates To
`optimize_pipeline`	Analyze a workflow trace and select a Pareto/MAB optimisation plan	`opt_meta.coordinate`
`optimization_report`	Get speed/cost/quality analysis without auto-apply	`opt_speed/cost/quality.analyze_and_recommend`
`optimization_stats`	Retrieve MAB arm statistics and eigen-behaviors	`opt_meta.get_intervention_stats`
`optimization_apply`	Apply a selected intervention plan with CSF gate and trace validation	`opt_meta.apply_plan`

Example: optimize_pipeline

{
  "trace": {
    "steps": [
      {"name": "search", "kind": "external_api", "latency_ms": 1200, "cost_usd": 0.003},
      {"name": "summarize", "kind": "llm_call", "latency_ms": 4500, "cost_usd": 0.08}
    ]
  },
  "auto_apply": false
}

Returns speed/cost/quality reports, Pareto frontier, selected intervention plan, and learning statistics when available.

Category 9 -- MCP Lifecycle (5)¶

Available on Builder only. Dynamic loading, unloading, and management of MCP sub-servers.

Tool	Description	Component
`nav_mcp_load`	Load an MCP sub-server for a component	`navigator`
`nav_mcp_unload`	Unload an active MCP sub-server	`navigator`
`nav_mcp_list_active`	List currently loaded MCP sub-servers	`navigator`
`nav_mcp_evict_idle`	Evict idle MCP sub-servers to free resources	`navigator`
`nav_mcp_registry`	View the full MCP sub-server registry	`navigator`

Category 10 -- Bias Evolution (43)¶

Available on premium access. (Trial accounts no longer reach this category: since 2026-08-16 the trial is the Researcher tool set, not Builder.) Bias observation, evaluation, and evolution with grounding checks. Mutating tools (bias_evolve, bias_apply, bias_rollback and the matching bias_mcp_* tools) also require a confirm_token that matches BIAS_EVOLUTION_CONFIRM_TOKEN.

Primary tools (7)¶

Tool	Description	Component
`bias_snapshot`	Capture the current bias state	`adapt_bias`
`bias_observe`	Record a bias observation	`adapt_bias`
`bias_evaluate`	Evaluate bias against principles	`adapt_bias`
`bias_evolve`	Evolve bias configuration	`adapt_bias`
`bias_apply`	Apply a bias evolution	`adapt_bias`
`bias_rollback`	Roll back to a previous bias state	`adapt_bias`
`bias_ground_check`	Verify bias against grounding sources	`adapt_bias`

Extended MCP operations (29)¶

The standalone adapt_bias MCP block exposes additional operations for deep bias management:

Tool	Description
`bias_mcp_snapshot`	Capture bias state (MCP block)
`bias_mcp_observe`	Record observation (MCP block)
`bias_mcp_understand`	Understand current bias landscape
`bias_mcp_evaluate`	Evaluate bias (MCP block)
`bias_mcp_evolve`	Evolve bias (MCP block)
`bias_mcp_apply`	Apply evolution (MCP block)
`bias_mcp_rollback`	Rollback (MCP block)
`bias_mcp_ground_check`	Ground check (MCP block)
`bias_mcp_list_versions`	List all bias versions
`bias_mcp_get_version`	Get a specific bias version
`bias_mcp_compare_versions`	Compare two bias versions
`bias_mcp_query_memory`	Query bias memory store
`bias_mcp_ingest_observation`	Ingest external observation
`bias_mcp_check_principle`	Check against a specific principle
`bias_mcp_list_principles`	List all bias principles
`bias_mcp_grounding_report`	Generate grounding report
`bias_mcp_self_heal`	Trigger self-healing cycle
`bias_mcp_health_check`	Health check for bias system
`bias_mcp_reset`	Reset bias state
`bias_mcp_export_state`	Export bias state
`bias_mcp_import_state`	Import bias state
`bias_mcp_clear_history`	Clear bias history
`bias_mcp_info`	Get bias system info
`bias_mcp_status`	Get bias system status
`bias_mcp_search`	Search bias records
`bias_mcp_dashboard`	Get bias dashboard data
`bias_mcp_promote`	Promote a bias version to active
`bias_mcp_stage_status`	Get bias staging status
`bias_mcp_compare_runs`	Compare bias evolution runs

Internal aliases

The 7 primary tools are also available with _bias_ prefix (e.g., _bias_snapshot) for internal pipeline use. These are counted separately in the total but behave identically.

Category 11 -- Self-Training (7)¶

Available on Builder only. Component maturity auditing, scoring, and promotion through the stability tiers.

Tool	Description	Component
`self_training_audit`	Audit a single component for maturity	`self_training`
`self_training_audit_all`	Audit all components in bulk	`self_training`
`self_training_scorecard`	Generate a maturity scorecard	`self_training`
`self_training_promote`	Promote a component to a higher stability tier	`self_training`
`self_training_flag`	Flag a component for review	`self_training`
`self_training_rubric`	Get the rubric for a specific tier	`self_training`
`self_training_rubrics`	List all available rubrics	`self_training`

Category 12 -- Tier Manager (5)¶

Available on Builder only. Practopoiesis-inspired tier advancement for component autonomy.

Tool	Description	Component
`tier_get_status`	Get current tier status for a component	`tier_manager`
`tier_propose_advance`	Propose advancing a component to the next tier	`tier_manager`
`tier_confirm_advance`	Confirm a tier advancement proposal	`tier_manager`
`tier_reject_advance`	Reject a tier advancement proposal	`tier_manager`
`tier_cap`	Set a maximum tier cap for a component	`tier_manager`

Note

tier_get_status is also available on Researcher plans for read-only access.

Category 13 -- Learning Layer (8)¶

Available on Builder only. Dataset construction, LLM-backed harness training/evaluation, and theory extraction for self-improving components.

Tool	Description	Component
`learning_build_dataset`	Build a training dataset from component traces	`learning_layer`
`learning_train`	Train a model on component data	`learning_layer`
`learning_evaluate`	Evaluate a trained model	`learning_layer`
`learning_get_theories`	Get extracted theories from training	`learning_layer`
`learning_get_artifacts`	Get training artifacts (models, datasets)	`learning_layer`
`learning_get_status`	Get learning pipeline status	`learning_layer`
`learning_export_harness`	Export a test harness for a trained model	`learning_layer`
`learning_import_harness`	Import an external test harness	`learning_layer`

Learning-layer caveats

learning_train and learning_evaluate can call the configured live LLM backend on every dataset row. Run a small smoke dataset first, set explicit budgets/timeouts, and run the gated smoke test on the deployment machine before a customer pilot:

$env:G6_WORKSPACE="C:\path\to\pilot-workspace"
$env:G6_LEARNING_LAYER_LIVE_SMOKE="1"
python -m pytest -q tests/mvp/learning_layer/test_live_llm_smoke.py

Use the same model/provider, credentials, workspace, and environment that will serve the pilot. If the smoke test fails, the live backend path is not launch-ready even if the mocked unit suite passes.

Use normalized_exact only for short deterministic answers. For open-ended outputs, use contains, token_f1, or rubric/llm_judge with concrete pass criteria. Rubric judging costs an extra LLM call per row and should be reviewed on borderline cases.

Training improves harness artifacts, prompts, theories, and generated skills; it does not fine-tune model weights and does not guarantee lift. Validate exported harnesses on held-out or frozen examples before using them in a production workflow.

Category 14 -- Patch Service (5)¶

Available on Builder only (read-only bug reporting available on Researcher). Patch management, application, and rollback.

Tool	Description	Component
`patch_report_bug`	Report a bug against any component	`patch_client`
`patch_check_status`	Check the status of a reported bug	`patch_client`
`patch_apply`	Apply a patch to a component	`patch_server`
`patch_list`	List available patches	`patch_server`
`patch_rollback`	Roll back an applied patch	`patch_server`

Note

patch_report_bug and patch_check_status are also available on Researcher plans.

Category 15 -- Clean Room (3)¶

Available on Builder only. Isolated component testing and bypass configuration.

Tool	Description	Component
`get_component_spec`	Get the specification for a component	`clean_room`
`configure_component_bypass`	Configure a component bypass for testing	`clean_room`
`list_bypasses`	List all active component bypasses	`clean_room`

Rate limiting¶

Rate limits are enforced per license key. When exceeded, the response includes a Retry-After value in seconds.

Tier	Requests / Minute
Free Trial	10
Researcher	100
Builder	500

Rate limits are configurable via environment variables: MCP_RATE_LIMIT_FREE, MCP_RATE_LIMIT_BASIC (Researcher), MCP_RATE_LIMIT_PREMIUM (Builder).