Skip to content

MCP Tools Reference

The Trurlic MCP server exposes twelve tools over stdio transport. Six are read-only (concurrent access, read lock only) and six are mutating (exclusive access, write lock + file lock).

Compute the workflow step for a component and return the next action. The orchestration hub — call before acting on a component and after completing each action. The returned ready field is the only signal that implementation can proceed.

ParameterTypeRequiredDescription
componentstringYesComponent name (kebab-case) or "project"
task_typeenumNonew_component, feature, fix, learn, review, harden, bootstrap. Inferred from graph state if omitted.
taskstringNoTask context passed to design prompts
step_evidenceobjectNoEvidence of user involvement for completed steps. Keys are step names, values are evidence strings. Gated steps require at least 20 bytes.

Returns: { step, action, requires_user_input, ready }.

Get the architectural context for a component. Returns decisions, project-wide rules, related decisions from connected components, and an authoritative brief for coding agents.

ParameterTypeRequiredDescription
componentstringYesComponent name or "project"
taskstringNoCurrent coding task description
depthenumNo"full" (default) or "constraints" (60–70% fewer tokens)

Check whether a pattern or approach is covered by existing decisions. Returns matching decisions sorted by relevance, or suggested_component if not covered.

ParameterTypeRequiredDescription
descriptionstringYesPattern or approach to check

Full architectural overview: components, connections, decision counts, patterns, and project-wide decisions. Lists components with zero decisions as needing design.

No parameters.

Full graph integrity check. Same validation as trurlic check.

No parameters.

Get the prompt for a specific workflow step. Called as directed by advance. Returns system instructions, component context, and step metadata.

ParameterTypeRequiredDescription
componentstringYesComponent name or "project"
stepenumYesWorkflow step name (see Workflow Engine)
taskstringNoTask context
task_typeenumNoTask type for variant prompts

All write tools validate the full graph before committing. Invalid mutations are refused with a clear error.

Record a single architectural decision. Validates all edges before writing. Atomic commit. Returns the decision name, path, warnings, and any detected pattern opportunities.

ParameterTypeRequiredDescription
componentstringYesComponent name or "project"
choicestringYesConcise decision title
reasonstringYesReasoning behind the decision
attributionenumYes"user" or "agent"
alternativesstring[]NoRejected options with reasons
depends_onstring[]NoDecision names this depends on
constrainsstring[]NoDecision names this constrains
tagsstring[]NoCategorical tags
supersedesstringNoDecision name being replaced

Record a pattern — a synthesis of multiple decisions into a reusable rule. Requires at least 2 decisions.

ParameterTypeRequiredDescription
namestringYesHuman-readable pattern name
descriptionstringYesWhat this pattern means
decisionsstring[]YesDecision names (minimum 2, must all exist)
componentsstring[]NoComponent names (inferred if omitted)
tagsstring[]NoCategorical tags

Modify an existing decision. Two modes: amend edits in place (typo, clarification); supersede creates a new decision replacing the old one (substantive change).

ParameterTypeRequiredDescription
namestringYesExisting decision name
modeenumYes"amend" or "supersede"
choicestringNoNew choice text
reasonstringNoNew reason text

At least one of choice or reason must be provided.

Remove a decision with cascade awareness. Refuses if other decisions depend on it or a pattern would shrink below 2 members. Reports affected patterns and edges.

ParameterTypeRequiredDescription
namestringYesDecision filename (without .toml)

Add a new component to the architecture graph.

ParameterTypeRequiredDescription
namestringYesKebab-case component name
descriptionstringNoOne-line description

Add a directional connection between two components. Represents data or control flow.

ParameterTypeRequiredDescription
fromstringYesSource component name
tostringYesTarget component name

The MCP dispatch uses tool classification to choose lock granularity:

  • Read tools (advance, get_context, check_pattern, get_architecture, validate_consistency, get_step_prompt) acquire a shared read lock — concurrent access, no blocking.
  • Write tools (record_decision, record_pattern, remove_decision, update_decision, add_component, add_connection) acquire an exclusive write lock plus a file lock — serialized access with full validation before commit.

Unknown tool names return an error response with isError: true.