Atlas / MCP servers / postrv / Narsil

NarsilBLOCK

Name: Narsil
Author: postrv

mcp/postrv/narsil

MCP server by postrv · code intelligence · mcp · mcp server · Audited 2026-10-06

⚑ Report / re-scan Source ↗

Rust MCP server for comprehensive code intelligence - 90 tools, 32 languages, security scanning, call graphs, and more

Verdict

BLOCK

Grade

Trust score

53 /100

Exposed tools

96 84r · 12w · 0d

Transport

—

License

Apache-2.0

Stars

184

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

The blazing-fast, privacy-first MCP server for deep code intelligence

[](LICENSE-MIT) [](https://www.rust-lang.org) [](https://github.com/postrv/narsil-mcp) [](https://modelcontextprotocol.io)

A Rust-powered MCP (Model Context Protocol) server providing AI assistants with deep code understanding through 90 specialized tools.

Why narsil-mcp?

Key Features

Code Intelligence - Symbol extraction, semantic search, call graph analysis
Neural Semantic Search - Find similar code using embeddings (Voyage AI, OpenAI)
Security Analysis - Taint analysis, vulnerability scanning, OWASP/CWE coverage
Supply Chain Security - SBOM generation, dependency auditing, license compliance
Advanced Analysis - Control flow graphs, data flow analysis, dead code detection

Why Choose narsil-mcp?

Written in Rust - Blazingly fast, memory-safe, single binary (~30MB)
Tree-sitter powered - Accurate, incremental parsing for 32 languages
Zero config - Point at repos and go
MCP compliant - Works with Claude, Cursor, VS Code Copilot, Zed, and any MCP client
Privacy-first - Fully local, no data leaves your machine
Parallel indexing - Uses all cores via Rayon
Smart excerpts - Expands

Read from source at commit c95b55984ca6OBSERVED · 2026-10-06

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add narsil-mcp --env API_KEY=${API_KEY} --env DATABASE_PASSWORD=${DATABASE_PASSWORD} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "narsil-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "API_KEY": "${API_KEY}",
        "DATABASE_PASSWORD": "${DATABASE_PASSWORD}"
      }
    }
  }
}

Exposed tools (96)

84 read · 12 write · 0 destructive.

Tool	Risk	Description
`add_remote_repo`	write	Add a remote GitHub repository for indexing. Clones the repo to a temporary location.
`check_cwe_top25`	read	Scan for CWE Top 25 Most Dangerous Software Weaknesses including buffer overflows, injection, improper input validation.
`check_dependencies`	read	Check project dependencies for known vulnerabilities using the OSV (Open Source Vulnerabilities) database. Returns CVE/GHSA IDs and recommended upgrades.
`check_licenses`	read	Analyze dependency licenses for compliance issues. Detects copyleft licenses, unknown licenses, and license compatibility problems.
`check_owasp_top10`	read	Scan specifically for OWASP Top 10 2021 vulnerabilities including injection, broken auth, XSS, SSRF, etc.
`check_type_errors`	read	Find potential type errors in Python/JavaScript/TypeScript code without running mypy/tsc. Detects type mismatches, undefined variables, etc.
`discover_repos`	read	Auto-discover repositories in a directory by detecting VCS roots and project markers
`explain_vulnerability`	read	Get detailed explanation of a security vulnerability type including examples, references, and remediation guidance.
`export_ccg`	read	Export all CCG layers as a bundle to a directory. Generates manifest.json, architecture.json, symbol-index.nq.gz.b64, and full-detail.nq.gz.b64.
`export_ccg_architecture`	read	Export CCG Layer 1 architecture (~10-50KB JSON-LD) with module hierarchy, public API, and dependencies.
`export_ccg_full`	read	Export CCG Layer 3 full detail (~1-20MB N-Quads gzipped) with complete RDF dataset including imports and security findings.
`export_ccg_index`	read	Export CCG Layer 2 symbol index (~100-500KB N-Quads gzipped) with all symbols and call graph edges.
`export_ccg_manifest`	read	Export CCG Layer 0 manifest to a file. Returns content or writes to specified path.
`find_call_path`	read	Find the call path between two functions. Requires --call-graph flag.
`find_calls`	read	Find what a function calls and what calls it
`find_circular_imports`	write	Detect circular import dependencies in the codebase. Returns all cycles with the files involved.
`find_dead_code`	read	Find unreachable code blocks in a function or file using control flow analysis.
`find_dead_stores`	read	Find variable assignments that are never read (dead stores).
`find_functions`	read	Find all functions in a repository
`find_functions_by_name`	read	Find functions matching a name pattern
`find_imports`	write	Find import relationships for a file
`find_injection_vulnerabilities`	read	Find injection vulnerabilities (SQL injection, XSS, command injection, path traversal) using taint analysis.
`find_references`	read	Find all references to a symbol across the codebase
`find_security_findings`	read	Find security findings at or above a severity threshold
`find_semantic_clones`	read	Find semantically similar code (Type-3/4 clones) using neural embeddings. Detects code that does the same thing with different implementation.
`find_similar_code`	read	Find code similar to a given snippet using TF-IDF embeddings. Good for finding duplicate or related code patterns.
`find_similar_to_symbol`	read	Find code similar to a specific symbol (function, class, etc.). Useful for finding related implementations or potential duplicates.
`find_symbol_usages`	read	Find all usages of a symbol across files, including imports and re-exports. Cross-language aware for JS/TS projects.
`find_symbols`	read	Find data structures (structs, classes, enums, interfaces) and functions/methods in a repository. Supports filtering by type and name pattern.
`find_symbols_in_file`	read	Find all symbols defined in a specific file
`find_uninitialized`	read	Find variables that may be used before being initialized.
`find_unused_exports`	read	Detect exported symbols never imported by other files in repo. Cross-file analysis using import graph. Configurable to exclude public API surface.
`find_upgrade_path`	read	Find safe upgrade paths for vulnerable dependencies. Shows which versions fix known vulnerabilities and whether upgrades have breaking changes.
`generate_sbom`	read	Generate a Software Bill of Materials (SBOM) for a project. Supports CycloneDX and SPDX formats. Parses Cargo.toml, package.json, requirements.txt, and go.mod.
`get_blame`	read	Get git blame information for a file. Requires --git flag.
`get_branch_info`	read	Get current branch name and repository status. Requires --git flag.
`get_call_graph`	read	Get the call graph for a repository or specific function. Requires --call-graph flag.
`get_callees`	read	Find functions called by a given function. Requires --call-graph flag.
`get_callers`	read	Find functions that call a given function. Requires --call-graph flag.
`get_ccg_access_info`	read	Get information about CCG access tiers and permissions. Explains the Triple-Heart Model and WebACL configuration.
`get_ccg_acl`	read	Generate WebACL access control document for CCG layers. Supports Triple-Heart Model (public/authenticated/private tiers).
`get_ccg_manifest`	read	Get CCG Layer 0 manifest (~1-2KB JSON-LD) with repository identity, symbol counts, languages, and security summary. Always fits in AI context window.
`get_chunk_stats`	read	Get statistics about code chunks in a repository.
`get_chunks`	read	Get AST-aware code chunks for a file with symbol context.
`get_code_graph`	write	Get graph visualization data (call graph, import graph, symbols). HTTP-only tool, not available via MCP.
`get_commit_diff`	write	Get the diff for a specific commit. Requires --git flag.
`get_complexity`	read	Get complexity metrics (cyclomatic, cognitive) for a function. Requires --call-graph flag.
`get_contributors`	read	Get contributors to a file or repository. Requires --git flag.
`get_control_flow`	read	Get the control flow graph (CFG) for a function, showing basic blocks, branches, and loops.
`get_data_flow`	read	Get data flow analysis for a function, showing variable definitions and uses.
`get_dependencies`	read	Analyze dependencies and imports for a file or module
`get_embedding_stats`	read	Get statistics about the embedding index.
`get_excerpt`	read	Extract code excerpts around specific lines with intelligent context expansion. Automatically expands to function/class boundaries when enabled.
`get_export_map`	read	Get the export map for a file or module showing all exported symbols and their types.
`get_file`	read	Get the contents of a specific file with optional line range
`get_file_history`	write	Get git commit history for a file. Requires --git flag.
`get_function_hotspots`	read	Find highly connected functions (potential refactoring targets) based on call graph analysis. Requires --call-graph flag.
`get_hotspots`	read	Find code hotspots - files with high churn and complexity. Requires --git flag.
`get_hover_info`	read	Get hover information (type info, documentation) for a symbol at a specific position. Enhanced with LSP when available.
`get_import_graph`	write	Build and analyze the import/dependency graph for a codebase. Shows which files import which other files, helps identify circular dependencies.
`get_incremental_status`	read	Get status of incremental indexing including Merkle tree root hash, file counts, and change statistics.
`get_index_status`	read	Get status of the search index and enabled features. Shows which optional features are enabled (--git, --call-graph, --persist, --watch) and index statistics.
`get_metrics`	read	Get performance metrics including tool execution times, indexing statistics, and server uptime
`get_modified_files`	read	Get list of modified files in the working tree. Requires --git flag.
`get_neural_stats`	read	Get statistics about the neural embedding index. Requires --neural flag.
`get_project_structure`	read	Get the directory structure and key files of a repository. Returns a tree view with file types and sizes.
`get_reaching_definitions`	read	Get reaching definitions analysis - which variable assignments reach each point in the code.
`get_recent_changes`	read	Get recent commits across the repository. Requires --git flag.
`get_remote_file`	read	Fetch a specific file from a remote GitHub repository via API (no clone needed).
`get_security_summary`	read	Get a comprehensive security summary for a repository including vulnerability counts and risk assessment.
`get_symbol_definition`	read	Get the full definition of a symbol with surrounding context. Returns the source code with line numbers.
`get_symbol_history`	read	Get commits that modified a specific symbol/function. Requires --git flag.
`get_taint_sources`	read	List all identified taint sources (user inputs, file reads, network data) in the codebase.
`get_type_info`	read	Get precise type information for a symbol. Requires LSP to be enabled.
`get_typed_taint_flow`	read	Enhanced taint analysis with type information. More precise than untyped taint tracking, combines data flow with type inference.
`go_to_definition`	read	Find the definition location of a symbol at a specific position. Enhanced with LSP when available.
`hybrid_search`	read	Perform hybrid search combining BM25 keyword search with TF-IDF semantic similarity using Reciprocal Rank Fusion (RRF).
`import_ccg`	write	Import a CCG layer from URL or local file. Supports JSON-LD (L0/L1) and gzipped N-Quads (L2/L3) formats.
`import_ccg_from_registry`	write	Import all CCG layers from the codecontextgraph.com registry for a repository.
`infer_types`	read	Infer types for variables in a Python/JavaScript/TypeScript function. Shows what types flow through the code without running external type checkers.
`list_remote_files`	read	List files in a remote GitHub repository via API (no clone needed). Rate limited without GITHUB_TOKEN.
`list_repos`	read	List all indexed repositories with metadata (path, language breakdown, file count)
`list_sparql_templates`	read	List available SPARQL query templates for common code intelligence patterns. Requires --graph flag.
`neural_search`	read	Search code using neural semantic embeddings. Finds semantically similar code even with different variable names. Requires --neural flag and EMBEDDING_API_KEY.
`query_ccg`	read	Query CCG Layer 3 using SPARQL. Enables rich semantic queries against the full code context graph.
`reindex`	write	Trigger re-indexing of a repository or all repositories
`run_sparql_template`	write	Execute a predefined SPARQL query template with parameters. Requires --graph flag.
`scan_security`	read	Scan repository for security issues using the security rules engine. Detects vulnerabilities, secrets, crypto issues, and more.
`search_chunks`	read	Search over AST-aware code chunks with symbol context.
`search_code`	read	Semantic and keyword search across code. Returns ranked excerpts with surrounding context.
`semantic_search`	read	BM25-ranked semantic search with code-aware tokenization. Better than simple text search for natural language queries.
`sparql_query`	write	Execute a SPARQL query against the RDF knowledge graph. Supports SELECT and ASK queries with timeout and result limits. Requires --graph flag.
`suggest_fix`	read	Get suggested fixes for a specific security finding.
`trace_taint`	read	Trace how tainted data flows from a source location through the code.
`validate_repo`	read	Validate that a path is a valid repository and can be indexed
`workspace_symbol_search`	read	Fuzzy search for symbols across the entire workspace. Uses trigram matching for typo-tolerant search.

Trust audit

BLOCKgrade F · trust 53/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	PASS
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (14 observation(s))
Network: declared (23 observation(s))
Shell: declared (14 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

CRITICALHard-coded secrets · secret.private_key · CWE-798, CWE-321

rules/secrets.yaml:230

- '-----BEGIN RSA PRIVATE KEY-----'

CRITICALHard-coded secrets · secret.private_key · CWE-798, CWE-321

rules/secrets.yaml:231

- '-----BEGIN PRIVATE KEY-----'

CRITICALHard-coded secrets · secret.private_key · CWE-798, CWE-321

rules/secrets.yaml:232

- '-----BEGIN EC PRIVATE KEY-----'

CRITICALHard-coded secrets · secret.private_key · CWE-798, CWE-321

rules/secrets.yaml:233

- '-----BEGIN OPENSSH PRIVATE KEY-----'

CRITICALHard-coded secrets · secret.private_key · CWE-798, CWE-321

rules/secrets.yaml:234

- '-----BEGIN DSA PRIVATE KEY-----'

CRITICALHard-coded secrets · secret.stripe · CWE-798, CWE-321

test-fixtures/security/vulnerable.kt:40

val apiKey = "sk_live_1234567890abcdefghij" // BAD: Hardcoded key

HIGHCode injection · code.deserialize · CWE-78, CWE-94, CWE-95

src/security_rules.rs:2484

data = pickle.loads(user_input)

Why it matters. deserialises untrusted bytes into live objects

Fix. use json or yaml.safe_load

HIGHCode injection · code.deserialize · CWE-78, CWE-94, CWE-95

test-fixtures/security/vulnerable.php:37

$obj = unserialize($data); // BAD: Unserialize user input

Why it matters. deserialises untrusted bytes into live objects

Fix. use json or yaml.safe_load

HIGHCode injection · code.deserialize · CWE-78, CWE-94, CWE-95

test-fixtures/security/vulnerable.php:38

return unserialize($_GET['obj']); // BAD

Why it matters. deserialises untrusted bytes into live objects

Fix. use json or yaml.safe_load

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/taint/patterns.rs:697

"exec(".to_string(),

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/taint/patterns.rs:699

"Exec(".to_string(),

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/taint/patterns.rs:720

"exec(".to_string(),

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/taint/patterns.rs:783

"eval(".to_string(),

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/taint/patterns.rs:784

"exec(".to_string(),

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

.claude/hooks/security-filter.sh:77

"$HOME/.ssh/"