Machine-readable manifesthttps://ai.primesync.me/api/tool-manifest
What you can build
- Build strategies
Turn your trading rules into candidate Pine Script or Python code.
- Create indicators
Build custom chart indicators using the logic you choose.
- Add alerts
Request alert code, then create, activate and test your alerts in TradingView.
- Fix or convert code
Repair scripts or convert indicators and strategies, with changes limited to what you approve.
- Set strategy Properties
Choose P0–P4 or Custom for account size, position sizing, costs and execution.
- Check your code
Review logic, repainting, timing, sizing and costs through static code checks.
- Research improvements
Turn supplied research into ideas and controlled experiments for your strategy.
- Analyze backtests
Review TradingView or Python trade exports across your chosen assets and timeframes.
- Run Monte Carlo tests
Explore possible drawdowns, losing streaks and uncertainty using your trade results.
- Continue a project
Resume with a project summary or optional saved preferences and strategy summaries.
See the alert setup guide for what to request from Studio and what to configure in TradingView.
Exact MCP tools
Compute-only result + metadata audit entry
Research & Build Engine
primesync_research_builder- Permissions
- strategies:build · strategies:validate
- Reads
- Only the goal, code, visual reference, client-supplied strategy ID and research ledger, and TradingView CSV data supplied in the current tool call.
- Returns
- A Strategy Completion Map; structured strategy profile; focused source queries; caller-attested and rejected evidence; transparent relevance scores; isolated source-linked hypotheses and candidate experiments; protected logic; missing evidence; validation and overfitting controls; integrity checks; and a returned research ledger. Independently verified sources are explicitly empty because the MCP server does not retrieve or authenticate papers.
- External effects
- Writes one metadata-only audit event (tool, scopes, client, time, status). It does not save code, execute submitted code, call brokers, publish scripts, or access TradingView accounts.
Compute-only result + metadata audit entry
Backtest Importer
primesync_import_backtest- Permissions
- strategies:validate
- Reads
- One supplied TradingView List of Trades CSV or Python closed-trade CSV plus explicit platform, asset, timeframe, timezone, and mapping metadata.
- Returns
- Dataset audit, normalized chronological returns, a dataset fingerprint, candidate/code-version/content-hash mapping, blueprint lineage, an explicit evidenceStatus, staged warnings, and the next action. BACKTEST EVIDENCE ACCEPTED requires matching blueprint hashes, a clean static pass, user-reported compilation, a user-attested fresh export, confirmed costs, matching confirmed effective Properties, and confirmed clean mapping; final selection additionally requires complete reproducibility metadata and comparable realistic Properties. P0/P4 remain comparison-only.
- External effects
- Writes one metadata-only audit event. It does not contact TradingView, save the CSV, or execute strategy code.
Compute-only result + metadata audit entry
Strategy Builder
primesync_strategy_builder- Permissions
- strategies:build
- Reads
- The trader's idea, rules, market, timeframe, execution assumptions, costs, requested Pine/Python output, and any connected-AI code candidate returned for audit.
- Returns
- A top-of-response Strategy Completion Map followed by a bundled clarification, deterministic Pine candidate, or open connected-AI authoring contract for arbitrary complete Pine/Python architectures. Returned candidates receive blueprint-integrity and static audit findings.
- External effects
- Writes one metadata-only audit event. It does not place trades, save workspaces, or alter external files.
Compute-only result + metadata audit entry
Static Code Audit
primesync_audit_strategy- Permissions
- strategies:audit
- Reads
- The approved rule blueprint, its required Strategy Builder hash, and supplied Pine Script or Python code.
- Returns
- A staged STATIC PASS, STATIC REVIEW REQUIRED, or STATIC FAILED result plus a stable code-content fingerprint and visible findings for blueprint drift, higher-timeframe leakage, look-ahead, negative offsets, pivots, intrabar execution, risk sizing, partial exits, stop mutation, sessions, pyramiding, alert/arrow parity, costs, direction, and rule coverage. Compilation and runtime behavior remain explicitly unconfirmed.
- External effects
- Writes one metadata-only audit event. It analyzes text and returns corrections in the conversation without changing external files.
Compute-only result + metadata audit entry
Monte Carlo
primesync_monte_carlo- Permissions
- strategies:validate
- Reads
- Up to 20,000 chronological returns per candidate, original row counts, return unit, up to 100,000 simulations, comparison objective, optional account failure threshold, and caller-supplied compilation/export/cost/mapping lineage.
- Returns
- A standalone analytical result, backward-compatible provisional comparison, or fully gated final-candidate decision. Final mode requires complete fold, holdout, cost-stress, parameter-sensitivity, code-content, dataset, and parameter lineage; it reports reproducible percentiles, drawdown, losing-streak distribution, negative-result and optional failure-threshold probabilities, rejection reasons, and the exact unchanged preferred code.
- External effects
- Writes one metadata-only audit event. It does not build code, fetch market data, or execute trades.
Optional account-owned memory reads, writes, export, and deletion + metadata audit entry
Personal preferences & strategy context
primesync_user_context- Permissions
- memory:manage
- Reads
- Only this authenticated user's confirmed preferences and explicitly saved project summaries. Shared reviewer accounts are excluded.
- Returns
- Memory state, revision-bound preferences or project summaries, a portable export, or one focused project-selection question. Remembered evidence remains user-reported and is never verification.
- External effects
- With explicit consent, saves or deletes allowlisted preferences and summaries. Memory defaults off. Disable deletes both. No raw code, CSVs, conversations, or full research ledgers are automatically saved.
Typed result contracts
Every MCP tool publishes an outputSchema. The machine-readable manifest includes the complete JSON Schema definitions and example result objects so clients can parse imports, blueprints, drift checks, research ledgers, and Monte Carlo metrics without guessing.
primesync_user_contextRequired: status
primesync_research_builderRequired: answer · status · operatingMode · taskMode · objective · strategyCompletionMap · strategyProfile · researchQueries · callerAttestedSources · verifiedSources · independentSourceVerification · sourceVerificationBoundary · rejectedSources · sourceRelevanceScores · hypotheses · candidateExperiments · protectedLogic · missingEvidence · validationPlan · overfittingWarnings · researchStatus · nextAction · validationProcess · researchLedger
primesync_import_backtestRequired: answer · status · evidenceStatus · dataset · chronologicalTradeReturns · chronologicalNetPnl · tradeIds · normalization · validationStages · lineage · verificationWarnings · traceabilityWarnings · nextAction
primesync_strategy_builderRequired: answer · status · blueprintHash · completionChecklist · sourceOfTradingRules · hiddenDefaults
primesync_audit_strategyRequired: answer · status · verificationStage · compileStatus · backtestStatus · robustnessStatus · approvedBlueprintHash · codeBlueprintHash · codeContentHash · blueprintDrift · counts · findings
primesync_monte_carloRequired: answer · status · finishedResult · method · randomSeed · selectionObjective · workflowMode · finalityStatus · evidenceProvenance · preferredCandidate · preferredCode · preferredCodeContentHash · rejectedCandidates · candidateResults · historical · simulated · codeChangeRule
Continuity is client-managed
strategyId and researchLedger carry the approved baseline, experiment history, accepted or rejected decisions, evidence, and blocked repeat suggestions. They are returned to your MCP client but are not stored by PrimeSync. The client must send the complete current ledger again on each relevant research call. Without it, only an explicitly selected saved summary can provide remembered decisions. Optional memory keeps minimal summaries and rejected-experiment notes, not the full research ledger or code. Recalled evidence never replaces fresh validation.
