Featured

Deploy OpenClaw in 60 seconds β€” 20% off logoDeploy OpenClaw in 60 seconds β€” 20% off

Launch OpenClaw on Hostinger in about 60 seconds and keep your agent live 24/7. Our referral link gives you 20% off, no coupon code needed.

Launch on Hostinger β†’
Run your Hermes agent on Hostinger, fully managed logoRun your Hermes agent on Hostinger, fully managed

Launch Hermes on Hostinger in one click, fully managed, no VPS knowledge needed. Use code ZACAARON10 for 10% off.

Launch on Hostinger β†’
Crawl and scrape any site into clean data, 10% off logoCrawl and scrape any site into clean data, 10% off

Firecrawl crawls and scrapes any site into clean markdown for your agent. Get 1,000 free credits, and new users get 10% off their first purchase.

Try Firecrawl free β†’
Your own AI agent, running 24/7 with QwikClaw logoYour own AI agent, running 24/7 with QwikClaw

QwikClaw sets up and runs an always-on OpenClaw agent for you. One click, no config files, no server setup.

Deploy now β†’
One API to scrape, enrich, and extract the internet. logoOne API to scrape, enrich, and extract the internet.

Context.dev gives your agents a single API to scrape, enrich, and extract live web data β€” no proxies, no parsers, no maintenance.

Start building free β†’
SetupClaw: done-for-you OpenClaw for founders & exec teams logoSetupClaw: done-for-you OpenClaw for founders & exec teams

White-glove OpenClaw for founders and exec teams (4–50+ employees): we install, harden, integrate your tools, and maintain it β€” secured from day one.

Get it set up for you β†’
SEO data APIs for your agent, $1 free credit logoSEO data APIs for your agent, $1 free credit

DataForSEO gives your agent live access to SERP results, keyword data, backlinks, and on-page SEO data through one API. New accounts get a $1 credit, good for up to 20,000 keyword or backlink lookups.

Try DataForSEO free β†’
Reach 47,000+ AI builders

A flat monthly placement in front of developers actively installing AI tools. No lock-in, cancel anytime.

Advertise here β†’

Works with

Claude CodeClaude DesktopCursorVS CodeClineCodex CLIOpenClaw+ any MCP client

Install to Claude Code

This server doesn't publish a one-line install command. Follow the setup in the source repository.

Summary

roslyn-codelens-mcp MCP server](https://glama.ai/mcp/servers/MarcelRoozekrans/roslyn-codelens-mcp/badges/score.svg)](https://glama.ai/mcp/servers/MarcelRoozekrans/roslyn-codelens-mcp) #️⃣ 🏠 🍎 πŸͺŸ 🐧 - Roslyn-based MCP server with 22 semantic code...

README.md

Roslyn CodeLens MCP Server

![MCP Toplist](https://mcptoplist.com/server/io.github.MarcelRoozekrans%2Froslyn-codelens)

![NuGet](https://www.nuget.org/packages/RoslynCodeLens.Mcp) ![NuGet Downloads](https://www.nuget.org/packages/RoslynCodeLens.Mcp) ![npm](https://www.npmjs.com/package/roslyn-codelens-mcp) ![Build Status](https://github.com/MarcelRoozekrans/roslyn-codelens-mcp/actions) ![License](https://github.com/MarcelRoozekrans/roslyn-codelens-mcp/blob/main/LICENSE) ![Docs](https://marcelroozekrans.github.io/roslyn-codelens-mcp/) ![GitHub Sponsors](https://github.com/sponsors/MarcelRoozekrans)

A Roslyn-based MCP server that gives AI agents deep semantic understanding of .NET codebases β€” type hierarchies, call graphs, DI registrations, diagnostics, refactoring, and more.

<a href="https://glama.ai/mcp/servers/MarcelRoozekrans/roslyn-codelens-mcp"> <img width="380" height="200" src="https://glama.ai/mcp/servers/MarcelRoozekrans/roslyn-codelens-mcp/badge" alt="roslyn-codelens-mcp MCP server" /> </a>

<!-- mcp-name: io.github.MarcelRoozekrans/roslyn-codelens -->

---

Hosted deployment

A hosted deployment is available on Fronteir AI.

Features

  • find_implementations β€” Find all classes/structs implementing an interface or extending a class
  • find_callers β€” Find every call site for a method, property, or constructor
  • find_event_subscribers β€” Every += / -= site for an event symbol, with resolved handler and subscribe/unsubscribe tag
  • find_tests_for_symbol β€” List xUnit/NUnit/MSTest methods that exercise a production symbol; opt-in transitive walk through helpers
  • get_test_summary β€” Per-project inventory of test methods with framework, attribute kind, data-row count, location, and production symbols referenced
  • find_uncovered_symbols β€” Public methods and properties no test transitively reaches; sorted by cyclomatic complexity for prioritization
  • generate_test_skeleton β€” Emit a compilable test-class skeleton (as text) for a method or type. Auto-detects xUnit/NUnit/MSTest; surfaces constructor dependencies as TodoNotes; returns a suggested file path. Closes the loop with find_uncovered_symbols
  • get_type_hierarchy β€” Walk base classes, interfaces, and derived types
  • get_di_registrations β€” Scan for DI service registrations. Reads generic (AddSingleton<IFoo, Foo>()), single-generic, typeof pair and factory-lambda forms
  • get_instantiation_options β€” "How do I construct this type?" in one call: constructors with full parameter detail, static factory methods declared anywhere in the solution (including on a separate factory type), DI registrations, and required members. Pass fromProject to learn whether that project can actually reach each option β€” it honours InternalsVisibleTo, so it answers "can my test project call this internal constructor?"
  • get_project_dependencies β€” Get the project reference graph
  • get_symbol_context β€” One-shot context dump for any type
  • get_public_api_surface β€” Enumerate every public/protected type and member in production projects; flat, deterministically-sorted list suitable for API review or breaking-change baselines.
  • find_breaking_changes β€” Diff the current API against a baseline JSON or DLL; report removed members, kind changes, and accessibility changes with Breaking/NonBreaking severity.
  • find_reflection_usage β€” Detect dynamic/reflection-based usage
  • get_exception_flow β€” What exceptions can escape a method: walks callees depth-bounded, propagates each throw up through every enclosing try/catch, and reports what still escapes with its propagation path; metadata callees contribute their documented exceptions
  • find_throw_sites β€” Every place an exception type is thrown, optionally including derived types; rethrows flagged
  • find_catch_blocks β€” Every catch for a type, optionally via base clauses; flags filtered, rethrowing, and empty (swallowing) handlers
  • find_references β€” Find all references to any symbol (types, methods, properties, fields, events), each tagged with a kind (read, write, readwrite, invocation, method_group, object_creation, cast, type_check, typeof, base_type, type_constraint, type_argument, declaration, attribute, nameof, xml_doc) and reported per occurrence with a column; filter server-side with kinds (e.g. ["write","readwrite"] for mutation sites)
  • go_to_definition β€” Find the source file and line where a symbol is defined
  • get_method_source β€” Full declaration source (XML docs, attributes, signature, body β€” original formatting) for one or many members by name in a single call: methods (all overloads), constructors, properties, indexers, fields, events; per-item statuses (ok/notFound/ambiguous/metadata/unsupportedKind) so a batch never fails wholesale
  • resolve_stack_trace β€” Map a pasted .NET stack trace to file/line/symbol, undoing compiler name mangling (async/iterator state machines, lambdas, local functions, generic arity); handles inner-exception chains, log-prefixed lines, and Demystifier traces
  • get_diagnostics β€” List compiler errors, warnings, and Roslyn analyzer diagnostics
  • get_code_fixes β€” Get available code fixes with structured text edits for any diagnostic
  • search_symbols β€” Fuzzy workspace symbol search by name
  • get_nuget_dependencies β€” List NuGet package references per project
  • find_attribute_usages β€” Find types and members decorated with a specific attribute
  • find_obsolete_usage β€” Every [Obsolete] call site grouped by deprecation message and severity, errors first; for planning migrations
  • find_circular_dependencies β€” Detect cycles in project or namespace dependency graphs
  • check_architecture β€” Enforce layering rules you supply (forbid and allowOnly) against the real semantic type graph rather than using directives; violations are grouped per boundary with a reference count and example sites
  • get_complexity_metrics β€” Cyclomatic complexity, cognitive complexity and max nesting depth per member (methods, constructors, properties, indexers, operators). Cyclomatic counts paths and starts at 1; cognitive measures how hard the code is to follow and starts at 0. metric picks which one the threshold and sort use β€” cognitive is the better refactoring-priority signal, cyclomatic the better test-budget one
  • find_naming_violations β€” Check .NET naming convention compliance
  • find_async_violations β€” Sync-over-async, async void misuse, missing awaits, fire-and-forget tasks; per-violation report with severity
  • find_disposable_misuse β€” IDisposable/IAsyncDisposable instances not wrapped in using/await using/returned/assigned to field; severity error/warning per violation.
  • find_large_classes β€” Find oversized types by member or line count
  • find_god_objects β€” Types combining high size with high cross-namespace coupling; sharper signal than raw size for SRP violations
  • find_unused_symbols β€” Dead code detection via reference analysis. Auto-filters test methods (xUnit/NUnit/MSTest), MCP tool entry points, source-generator output, MEF-composed services, and interop-laid-out fields; filter counts surface in summary.filteredOut
  • get_project_health β€” Composite audit aggregating 7 quality dimensions per project (complexity, large classes, naming, unused symbols, reflection, async violations, disposable misuse) with counts and top-N hotspots inline
  • get_source_generators β€” List source generators and their output per project
  • get_generated_code β€” Inspect generated source code from source generators
  • inspect_external_assembly β€” Browse types, members, and XML docs from closed-source NuGet packages and referenced assemblies
  • peek_il β€” Decompile any method to ilasm-style IL bytecode from closed-source or generated assemblies
  • get_code_actions β€” Discover available refactorings and fixes at any position (extract method, rename, inline variable, and more)
  • apply_code_action β€” Execute any Roslyn refactoring by title, with preview mode (returns a diff before writing to disk)
  • rename_symbol β€” Solution-wide safe rename of a type or member via Roslyn's Renamer, with preview mode, conflict reporting, and a freshness check against on-disk edits
  • change_signature β€” Add, remove, and reorder a method's parameters and rewrite every call site; handles named/optional arguments, params and extension methods, and reports the overrides and interface implementations it cascaded to
  • list_solutions β€” List all loaded solutions and which one is currently active
  • set_active_solution β€” Switch the active solution by partial name (all subsequent tools operate on it)
  • load_solution β€” Load an additional .sln/.slnx at runtime and make it the active solution
  • unload_solution β€” Unload a loaded solution to free memory
  • rebuild_solution β€” Force a full reload of the analyzed solution
  • start_background_task β€” Queue a long-running tool (currently rebuild_solution) to run in the background; returns a taskId to poll
  • get_task_status β€” Get the current status, result, or error of a background task by its taskId
  • list_running_tasks β€” List background tasks running or completed within the last 5 minutes
  • trust_solution β€” Authorize a solution to run Roslyn analyzers (required before get_diagnostics with includeAnalyzers: true)
  • list_trusted_paths β€” Inspect the persistent trust store + session-trusted solutions
  • revoke_trust β€” Revoke a previously-granted trust for a solution path
  • analyze_data_flow β€” Variable read/write/capture analysis within a statement range (declared, read, written, always assigned, captured, flows in/out)
  • analyze_control_flow β€” Branch/loop reachability analysis within a statement range (start/end reachability, return statements, exit points)
  • analyze_change_impact β€” Show all files, projects, and call sites affected by changing a symbol β€” combines find_references and find_callers
  • get_type_overview β€” Compound tool: type context + hierarchy + file diagnostics in one call
  • analyze_method β€” Compound tool: method signature + callers + outgoing calls in one call
  • get_overloads β€” Every overload of a method/constructor (source + metadata) with full parameter and modifier detail in one call
  • get_extension_methods β€” Every extension member applicable to a type β€” including arrays, nullables and tuples β€” from the solution and referenced assemblies (so LINQ shows up), using Roslyn's own applicability rules; covers C# 14 extension blocks including properties and static members
  • get_operators β€” Every user-defined operator and conversion operator on a type (source + metadata) with kind, signature, parameters, and source location. Includes synthesized record equality and .NET 7+ checked variants
  • get_call_graph β€” Transitive caller/callee graph for a method, depth-bounded with cycle detection
  • get_file_overview β€” Compound tool: types defined in a file + file-scoped diagnostics in one call

Response shape

All list-returning tools wrap their results in a uniform envelope:

{
  "items": [ ... ],
  "totalCount": 142,
  "truncated": false,
  "limit": 500,
  "summary": { ... }
}

When truncated is true, the items are the top N by the tool's natural sort order (severity-first, worst-first, by-project, etc.) β€” usually that's exactly what you want. Raise limit only if the missing tail items matter for the task.

Tools that include a summary aggregate today:

  • get_diagnostics β€” { error, warning, info, hidden } counts
  • find_references β€” { byProject: { name: count }, byKind: { kind: count } }
  • find_callers, find_attribute_usages β€” { byProject: { name: count } }
  • search_symbols, find_reflection_usage β€” { byKind: {...} }
  • find_unused_symbols β€” { byKind, filteredOut: { testMethod, testContainer, mcpTool, generated, composition, interop } }
  • find_naming_violations β€” { byRule: {...} }
  • get_complexity_metrics β€” { max, avg, overThreshold, maxCognitive } (the first three describe the selected metric)

Single-object tools (get_type_overview, get_symbol_context, apply_code_action, etc.) return their bespoke shape directly β€” the envelope only wraps list-returning tools.

Error responses

When a tool can't proceed (symbol not resolved, solution not trusted, file not found, ambiguous match, etc.), the response is an isError: true content block carrying a structured JSON body:

{
  "code": "SolutionNotTrusted",
  "message": "Solution 'Foo.sln' is not trusted for analyzer execution. ...",
  "details": { "solutionPath": "C:\\Foo.sln" }
}

Error codes (switch on code to handle each):

  • SymbolNotFound β€” type / method / property could not be resolved.
  • SolutionNotTrusted β€” get_diagnostics or get_code_fixes requested analyzers but the solution hasn't been authorized via trust_solution.
  • AmbiguousMatch β€” set_active_solution / unload_solution matched multiple solutions; details.matches lists them.
  • FileNotFound β€” file path or baseline doesn't exist (or isn't in any loaded project).
  • ProjectNotFound β€” solution name didn't match any loaded solution.
  • InvalidArgument β€” caller-supplied input was malformed, unsupported, or out of range.
  • Internal β€” unexpected error not modeled above; message carries the underlying exception text.

Cancellation: the MCP framework's native cancellation is honored. Cancelling a tools/call request mid-flight terminates the operation; long-running tools (get_diagnostics with analyzers, get_code_actions, apply_code_action, get_code_fixes) check the token at hot-loop boundaries.

External Assemblies

Metadata-origin symbols (from NuGet packages and referenced assemblies) are first-class citizens:

  • Tier 1 β€” Navigation (find_references, find_callers, find_implementations): Accepts closed-source type and member names. Resolves them from assembly metadata and reports all source-level usage sites.
  • Tier 2 β€” Inspection (inspect_external_assembly): Browse namespaces, types, members, and XML doc comments from any referenced assembly without decompiling.
  • Tier 3 β€” IL (peek_il): Decompile a specific method to annotated ilasm-style IL using ICSharpCode.Decompiler β€” useful for understanding the internals of NuGet libraries.

Location-returning results include an Origin field (source or metadata) and an IsGenerated flag to distinguish hand-written code from closed-source or generated output.

Runtime configuration

  • ROSLYN_CODELENS_OPEN_PROJECT_TIMEOUT_SECONDS β€” per-project MSBuild load timeout (default 300). When a project exceeds this duration during workspace open, it's recorded as a SkippedProjects entry with kind: "Timeout" and the rest of the solution still loads. Useful when a legacy or malformed project wedges the BuildHost-net472 subprocess.

Security: Trust Model

get_diagnostics and get_code_fixes can load Roslyn analyzers β€” DLLs that execute in-process. To prevent untrusted analyzers from running automatically, this server uses a VS/Rider-style trust model:

  • Solutions passed on the CLI at startup are auto-trusted for the current session.
  • Other solutions must be explicitly trusted via the trust_solution MCP tool.
  • Analyzer DLLs must come from the user's NuGet global packages folder, the dotnet SDK install dir, or the solution's own bin/obj. Other paths are skipped.

Use the list_trusted_paths and revoke_trust tools to inspect and manage trust state. Persistent trust is stored at %APPDATA%\roslyn-codelens\trust.json.

See SECURITY.md for the full threat model.

Quick Start

npx (any MCP client)

{
  "mcpServers": {
    "roslyn-codelens": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "roslyn-codelens-mcp"]
    }
  }
}

The npm package ships no server code β€” it is a launcher that installs the RoslynCodeLens.Mcp .NET global tool at a matching version and execs it, so the .NET 10 SDK must be on PATH. Subsequent starts skip the install entirely and work offline.

VS Code / Visual Studio (via dnx)

Add to your MCP settings (.vscode/mcp.json or VS settings):

{
  "servers": {
    "roslyn-codelens": {
      "type": "stdio",
      "command": "dnx",
      "args": ["RoslynCodeLens.Mcp", "--yes"]
    }
  }
}

Claude Code Plugin

claude install gh:MarcelRoozekrans/roslyn-codelens-mcp

.NET Global Tool

dotnet tool install -g RoslynCodeLens.Mcp

Then add to your MCP client config:

{
  "mcpServers": {
    "roslyn-codelens": {
      "command": "roslyn-codelens-mcp",
      "args": [],
      "transport": "stdio"
    }
  }
}

Docker

Runs without a .NET SDK on the host; the solution is bind-mounted at /workspace.

docker build -t roslyn-codelens-mcp .
docker run -i --rm -v "$PWD:/workspace" roslyn-codelens-mcp

The mounted solution must be restored for MSBuildWorkspace to resolve its references, and tool output reports container paths rather than host paths β€” see docs/site/docs/getting-started/docker.md.

Usage

The server automatically discovers .sln files by walking up from the current directory. You can also pass one or more solution paths directly:

# Single solution
roslyn-codelens-mcp /path/to/MySolution.sln

# Multiple solutions β€” switch between them with set_active_solution
roslyn-codelens-mcp /path/to/A.sln /path/to/B.sln

When multiple solutions are loaded, use list_solutions to see what's available and set_active_solution("B") to switch context. The first path is active by default.

Performance

All type lookups use pre-built reverse inheritance maps, member indexes, and attribute indexes for O(1) access. Benchmarked on an i9-12900HK with .NET 10.0.7:

| Tool | Latency | Memory | |------|--------:|-------:| | go_to_definition | 2.1 Β΅s | 576 B | | find_implementations | 2.5 Β΅s | 720 B | | get_project_dependencies | 2.8 Β΅s | 1.5 KB | | get_type_hierarchy | 3.5 Β΅s | 1.3 KB | | find_circular_dependencies | 3.7 Β΅s | 2.7 KB | | get_symbol_context | 4.1 Β΅s | 1.0 KB | | get_source_generators | 16 Β΅s | 23 KB | | analyze_data_flow | 19 Β΅s | 1.6 KB | | find_attribute_usages | 72 Β΅s | 904 B | | get_generated_code | 78 Β΅s | 24 KB | | analyze_control_flow | 115 Β΅s | 14 KB | | inspect_external_assembly (summary) | 159 Β΅s | 35 KB | | find_large_classes | 265 Β΅s | 3.5 KB | | get_di_registrations | 478 Β΅s | 16 KB | | inspect_external_assembly (namespace) | 564 Β΅s | 259 KB | | find_reflection_usage | 705 Β΅s | 19 KB | | get_complexity_metrics | 781 Β΅s | 25 KB | | get_code_actions | 792 Β΅s | 54 KB | | get_file_overview | 797 Β΅s | 101 KB | | get_diagnostics | 822 Β΅s | 99 KB | | get_nuget_dependencies | 849 Β΅s | 48 KB | | get_public_api_surface | 885 Β΅s | 247 KB | | get_type_overview | 1.1 ms | 104 KB | | peek_il | 1.1 ms | 34 KB | | find_disposable_misuse | 3.5 ms | 286 KB | | find_uncovered_symbols | 3.8 ms | 224 KB | | search_symbols | 3.9 ms | 557 KB | | analyze_method | 5.8 ms | 333 KB | | find_async_violations | 7.0 ms | 335 KB | | find_tests_for_symbol (direct) | 8.4 ms | 396 KB | | find_callers | 10 ms | 337 KB | | find_tests_for_symbol (transitive) | 12 ms | 399 KB | | find_naming_violations | 15 ms | 788 KB | | find_unused_symbols | 23 ms | 1.0 MB | | find_references | 28 ms | 1013 KB | | analyze_change_impact | 33 ms | 1.3 MB | | Solution loading (one-time) | ~4.1 s | 16 MB |

Hot Reload

The server watches .cs, .csproj, .props, and .targets files for changes. When a change is detected, affected projects are lazily re-compiled on the next tool query β€” only stale projects and their downstream dependents are re-compiled, not the full solution.

Location-returning tools include an IsGenerated flag to distinguish source-generator output from hand-written code.

Requirements

  • .NET 10 SDK
  • A .NET solution with compilable projects

Project compatibility

The server analyses every project that MSBuildWorkspace can load under the .NET SDK runtime.

Supported: SDK-style projects (<Project Sdk="...">), any target framework β€” net48, net6.0, net8.0, net10.0, etc. .NET Framework targets work fine as long as the csproj uses the SDK-style format.

Skipped (with a warning, not a crash): legacy non-SDK-style projects (<Project ToolsVersion="..." xmlns="http://schemas.microsoft.com/developer/msbuild/2003">, typically .NET Framework projects authored in older versions of Visual Studio). These rely on Microsoft.Common.props imports from the .NET Framework MSBuild that ships with Visual Studio, which is not available in the .NET SDK MSBuild runtime.

When a solution contains legacy projects, the server:

  1. Loads every SDK-style project normally β€” all tools work for those.
  2. Skips each legacy project and records it in LoadedSolution.SkippedProjects.
  3. Surfaces the skipped list via list_solutions (the SkippedProjects array on each SolutionInfo) and in the return message of load_solution. Each entry includes the project name, kind (Legacy), and reason.

To analyse a legacy project, convert it to SDK-style format (see Microsoft's migration guide) or open the solution from a Visual Studio Developer Command Prompt so the full Visual Studio MSBuild is on PATH.

Development

dotnet build
dotnet test
dotnet run --project benchmarks/RoslynCodeLens.Benchmarks -c Release

Third-party licenses

License

MIT

See related servers & alternatives β†’

Related MCP servers

Browse all β†’

Related guides

Hand-picked reading to help you choose and use Developer Tools servers.