Skip to content

MCP Tools

SYNAPSEED exposes 25 tools via the Model Context Protocol. Tools are callable actions that the LLM can invoke.

PRIMARY TOOL

Start with ask — The intent router orchestrates all subsystems and provides the best response for any query. Only use low-level tools when you need direct access to specific subsystems.

hoist

Index a project directory and return its AST skeleton.

Parameters:

NameTypeRequiredDescription
pathstringNoDirectory to index (default: project root)

Returns: JSON with files_indexed, symbols_indexed, path.


lookup

Find a symbol by name across the entire project.

Parameters:

NameTypeRequiredDescription
namestringYesSymbol name to search for

Returns: Array of matching symbols with file, line range, kind, and signature.


scan

Scan text content for sensitive data (API keys, passwords, tokens, PII) and/or code vulnerability patterns (SQL injection, XSS, command injection, path traversal).

Parameters:

NameTypeRequiredDescription
contentstringYesText to scan
modestringNoScan mode: all (default — DLP + patterns), dlp (secrets only), patterns (code vulnerability patterns only)

Returns: CLEAN or ALERT with findings and redacted output. When mode is all, combines DLP and code pattern results.


check

Evaluate a shell command against the security policy.

Parameters:

NameTypeRequiredDescription
commandstringYesShell command to evaluate

Returns: ALLOWED or DENIED with reason.


blame

Get git blame/history for a file.

Parameters:

NameTypeRequiredDescription
filestringYesFile path (relative to project root)
start_lineintegerNoStart line (default: 1)
end_lineintegerNoEnd line (default: 50)

Returns: Array of blame entries with commit, author, timestamp, and message.


diagnose

Run a full diagnostic on the project.

Parameters: None.

Returns: Project state, git status, recent commits, and metrics.


consult

Consult the project's architecture policy.

Parameters:

NameTypeRequiredDescription
querystringYesArchitecture question

Returns: Architecture guidance from DNA configuration.


Search for code by concept using Tantivy.

Parameters:

NameTypeRequiredDescription
querystringYesSemantic search query
limitintegerNoMax results (default: 5)

Returns: Ranked search results with file, symbol, score, and context.


diagnostics

Get current compiler diagnostics from the background shadow compiler. Supports severity filtering.

Parameters:

NameTypeRequiredDescription
filestringNoFilter by file path
min_severitystringNoMinimum severity: info, warning (default), error

Returns: Errors and warnings with codes, messages, and suggestions. Returns CLEAN if no diagnostics match the filter.


analyze

Analyze the full history of a file.

Parameters:

NameTypeRequiredDescription
filestringYesFile path
start_lineintegerNoScope to start line
end_lineintegerNoScope to end line

Returns: Churn score, co-change patterns, semantic tags, risk assessment.


quickfix

Apply a compiler-suggested fix automatically.

Parameters:

NameTypeRequiredDescription
filestringYesFile with the error
error_codestringYesError code to fix

Returns: Success message or error. Only applies MachineApplicable suggestions.


ask

The Intent Router. Ask a natural-language question and get an orchestrated response.

Parameters:

NameTypeRequiredDescription
querystringYesNatural-language question
rawbooleanNoWhen true, inject the exact source code of discovered symbols into the prompt (Direct Symbol Injection). Default: false

Returns: Enriched context with diagnostics, history, code context, security status, and SID (Semantic Information Density) metric. When raw: true, includes verbatim source code between --- FILE: path (lines X-Y) --- / --- END --- delimiters. Output format adapts to the detected model tier (Atomic/Molecular/Galactic).


reset-telemetry

Clear all telemetry data from the OTLP receiver.

Parameters: None.

Returns: Confirmation with count of cleared spans and locations.


intent

Summarize the intent and direction of recent commits semantically. Groups commits by category (fix, feature, refactor, security, etc.) and extracts scope hints from conventional commit messages.

Parameters:

NameTypeRequiredDescription
limitintegerNoNumber of recent commits to analyze (default: 20)

Returns: Natural-language summary with category breakdown and JSON detail.


train

Evaluate Rust code in an isolated sandbox (The Gym). Compiles, tests, benchmarks, and optionally runs adversarial mutation testing, returning metrics and a composite score.

Parameters:

NameTypeRequiredDescription
sourcestringYesRust source code to evaluate (injected as lib.rs)
testsstringNoOptional test code (injected as tests/eval.rs). Use use eval_project::*; to import.
timeoutintegerNoMax evaluation time in seconds (default: 60)
fuzzbooleanNoEnable proptest fuzzing: auto-generate property tests for public functions (default: false)
adversarialbooleanNoEnable adversarial mutation testing: apply controlled mutations to measure test suite effectiveness (default: false)

Returns: Compilation status, test results, mutation score (if adversarial), and detailed diagnostics with composite score.


janitor

Run the Janitor: scan for clippy warnings and unused dependencies, generate validated fix proposals.

Parameters: None.

Returns: Findings and actionable proposals with UUIDs for janitor-fix.


janitor-fix

Apply a specific Janitor fix proposal. Dry-run by default — shows a preview of what would change. Set confirm: true to actually apply.

Parameters:

NameTypeRequiredDescription
proposal_idstringYesUUID of the proposal (from janitor)
confirmbooleanNoSet to true to apply. Default: false (preview only)

Returns: Preview diff (dry-run) or success/error message (confirmed). Automatically reverts if compilation breaks.


approve-fix

Apply a RepairOrchestrator auto-fix proposal. Dry-run by default — shows a preview. Set confirm: true to apply. The fix is verified with cargo check and auto-reverted on failure.

Parameters:

NameTypeRequiredDescription
proposal_idstringYesUUID of the auto-repair proposal
confirmbooleanNoSet to true to apply. Default: false (preview only)

Returns: Preview diff (dry-run) or success/error message (confirmed). Automatically reverts if compilation breaks.


architect

Analyze project structural health: dependency graph, coupling metrics, cycle detection, god objects, layer violations.

Parameters:

NameTypeRequiredDescription
refreshbooleanNoForce fresh analysis (default: use cached report)

Returns: Architecture score (A-F), violations, module metrics, and recommendations.


similar

Find code similar to a natural-language query using vector embeddings (cosine similarity). Requires search.embeddings: true in DNA config.

Parameters:

NameTypeRequiredDescription
querystringYesNatural-language description of code to find
top_kintegerNoNumber of results (default: 5)
min_similaritynumberNoMinimum cosine similarity threshold (default: 0.3)

Returns: Ranked results with file, symbol, similarity score, and code snippet.


oracle

Auto-repair drifted documentation. Updates version numbers, crate counts, and MCP tool/resource counts in README.md to match the actual codebase.

Parameters: None.

Returns: List of changes made (or "no drift detected").


verify_path

Verify whether a file path exists within the project. Prevents LLM hallucination of file names by providing ground truth.

Parameters:

NameTypeRequiredDescription
pathstringYesFile path to verify (relative to project root)

Returns: { exists, size_bytes, language, is_file }. Path traversal attempts are blocked.


analyze_binary

The Neural Decompiler. Analyze a compiled binary (ELF/Mach-O/PE) to extract symbols, strings, and behavioral intent.

Parameters:

NameTypeRequiredDescription
pathstringYesPath to the binary file
max_symbolsintegerNoMax symbols to return (default: 50)
max_stringsintegerNoMax strings to return (default: 30)

Returns: Analysis summary, behavioral tags, and top symbols/strings.


explain_dependency

Analyze a compiled Rust dependency by crate name.

Parameters:

NameTypeRequiredDescription
crate_namestringYesName of the crate to explain (e.g., "tokio")

Returns: Binary analysis of the crate's compiled artifact (rlib/dylib).


run_benchmark

Run a reproducible SCR evaluation suite.

Parameters:

NameTypeRequiredDescription
suite_pathstringYesPath to the JSONL benchmark suite
formatstringNoOutput format: summary (default) or json

Returns: Benchmark report with F1, SCR, and hallucination metrics.

Released under the Apache License 2.0.