signalint
Signalint is a read-only local MCP server for JavaScript/TypeScript diagnostics that runs linters/type checkers, clusters issues, and tracks loops.
ping: Verify the server is responsive (returns pong).
check_project: Run full-project diagnostics with Oxlint, tsc, and optionally Biome; cluster issues by root cause with priorities and warnings.
check_files: Run incremental diagnostics on specified files, using content-hash caching to skip unchanged files.
get_issue_detail: Fetch full details for a cluster or issue from the latest check; returns stale if no longer present.
get_loop_status: List diagnostic signatures currently looping or file-rule churn from accumulated session history.
All tools are read-only, modify no files, and require no authentication.
Optionally runs Biome to provide linting and formatting diagnostics, with support for Biome configuration and caching.
Runs the TypeScript compiler to provide type checking and diagnostics for JavaScript and TypeScript projects, with support for project references and incremental caching.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@signalint Run lint and type checking on this project"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Signalint
Signalint is a reliable, read-only-by-default diagnostic intelligence layer for AI coding agents (Claude Code, Cursor, Codex, VS Code) and CI workflows.
Instead of flooding an agent's context window with thousands of raw compiler lines, Signalint normalizes diagnostics across compilers, linters, and language servers into bounded, priority-clustered issue summaries, calculates verification deltas against explicit baselines, and detects oscillating fix loops.
Key Differentiators
13 Complete MCP Tools: From incremental checks and immutable snapshots to live diagnostic ingestion and journaled transactional fix application.
Verification Deltas: Exact arithmetic on introduced, resolved, and unchanged diagnostics with semantic identity matching that survives line shifts.
Immutable Snapshots & Concurrency Safety: Isolated in-memory snapshot store keyed by
checkIdeliminates race collisions during concurrent checks.Monorepo-Aware Planning: Unites package manifests (
pnpm-workspace.yaml) parsed via pure-JS YAML with glob negation exclusions and TypeScript Project References, ordering checks topologically.Multi-Source Provenance: Distinguishes compiler, linter, and editor/LSP diagnostics across JavaScript, TypeScript, Python (Ruff/Mypy), Rust (Clippy), and Go (GolangCI-Lint).
Portable Host Hooks: Native verify-after-change adapters for Claude Code, Cursor, Codex, and VS Code.
In-Memory Previews & In-Process Rollback: In-memory previews never touch disk; apply uses per-file adjacent temporary-file replacement, pre-write drift detection, and in-process best-effort rollback for multi-file writes (does not claim persistent crash recovery if aborted abruptly).
Related MCP server: agent-workspace-mcp
Quick Start
1. Requirements
Node.js >= 22.12.0 (utilizes native
node:sqlite).Your project installs its own diagnostic engines (
typescript,oxlint,eslint, and/or@biomejs/biome). Signalint detects project-local installations and ships no bundled engines.
2. Initialize in your repository
npx signalint initThis detects your project's linters/compilers, generates signalint.config.json, and configures your installed MCP clients.
3. Add to MCP Clients
Add Signalint to your MCP configuration (e.g. claude_desktop_config.json, .cursor/mcp.json):
{
"mcpServers": {
"signalint": {
"command": "npx",
"args": ["-y", "signalint-mcp"]
}
}
}Complete MCP Tools Catalog
Signalint exposes 13 strictly typed MCP tools over stdio:
Tool | Category | Safety Annotation | Description |
| Diagnostics | Read-only ( | Verifies server responsiveness and canonical project root path. |
| Diagnostics | Read-only ( | Full project/workspace scan across all enabled engines. |
| Diagnostics | Read-only ( | Fast incremental check on specific files using content-hash caching. |
| Diagnostics | Read-only ( | Retrieves full, unomitted issue records anchored to an immutable |
| Diagnostics | Read-only ( | Returns oscillating fix signatures and rule churn detected in the session. |
| Snapshots | Read-only ( | Retrieves complete state and metrics for an immutable check snapshot. |
| Deltas | Read-only ( | Computes verification delta between a baseline check and a current check. |
| Deltas | Read-only ( | Runs incremental check on edited files and diffs against a baseline check. |
| Live / LSP | Read-only ( | Ingests external LSP or compiler diagnostics into Signalint's clustering model. |
| Live / LSP | Read-only ( | Returns active buffered diagnostics with source provenance. |
| Transactions | Read-only ( | Stages proposed code fixes in-memory. Does not write to disk. |
| Transactions | Mutating ( | Per-file adjacent temp replacement, pre-write drift check, and in-process best-effort rollback. |
| Transactions | Read-only ( | Drops an in-memory fix transaction without touching disk. |
See docs/mcp-contract.md for full JSON Schemas and error codes.
CLI Commands
Signalint includes a comprehensive command-line interface:
# Run a project diagnostic scan:
npx signalint check .
npx signalint check src/ --format github # GitHub Actions annotations
# View session statistics and cache economics:
npx signalint stats
# Verify local engine availability and configuration:
npx signalint doctor
# Manage host verify-after-change hooks:
npx signalint hooks preview --runtime claude
npx signalint hooks install --runtime cursor --confirm
npx signalint hooks run --runtime claude --payload '<json-event>'Monorepo Support
Signalint natively understands multi-package monorepos:
Configure
"monorepoMode": "auto"insignalint.config.jsonto detectpnpm-workspace.yamlvia pure-JS YAML parsing with glob negation exclusions (!).Strictly differentiates local workspace packages (
workspace:*) from external npm registry dependencies.Builds a topological dependency graph uniting package manifests and TypeScript project references.
Caches whole-program TypeScript builds in isolated per-package directories (
.signalint/cache/tsc/<project-id>/).Changes to a shared package automatically invalidate dependent packages while keeping independent siblings cached.
See docs/monorepo.md for configuration and architecture.
Verify-After-Change Hooks
Automate post-edit verification without creating infinite agent loops:
Native adapters for Claude Code, Cursor, Codex, and VS Code.
Emits bounded context into prompt windows and blocks only on unresolved syntax/type errors.
See docs/hooks.md for setup instructions.
Documentation Index
License
MIT © 2026 Tran Trong Qui
Available Tools
13 toolsafter_edit_checkARead-onlyIdempotent
Runs incremental diagnostics on specified edited files and computes a verification delta against an optional baselineCheckId. Read-only; no files are written or modified. Use this after modifying files to verify whether edits introduced or resolved diagnostics relative to the baseline check.
Inputs: files (required), baselineCheckId. Returns: code, delta, engine, status, +16 more
apply_diagnostic_fixADestructive
Applies diagnostic modifications using per-file adjacent temporary-file replacement, pre-write drift detection, and in-process best-effort rollback for multi-file writes. Requires explicit confirm: true. If baselineCheckId is provided, it must be valid and active in the snapshot store; unknown or expired baselines abort immediately without modifying files. Does not provide persistent crash recovery if the process terminates abruptly.
Inputs: confirm (required), transactionId (required), baselineCheckId. Returns: code, delta, error, status, +4 more
check_filesARead-onlyIdempotent
Runs Oxlint and TypeScript (and optionally Biome) lint and type diagnostics on a specific list of files, using per-engine content-hash caching to skip unchanged files. Read-only; no files are written or modified. Use this for incremental checks after editing specific files; use check_project for a full project scan. The files parameter expects relative file paths (not glob patterns) within the project directory — absolute paths or paths outside the root return an error response. Caching is file-content-hash-based: a file is re-checked only when its content or the engine's config file (e.g., .oxlintrc, tsconfig.json) has changed since the last call, not based on git status. TypeScript is a whole-program engine: it re-runs whenever any TypeScript file in the request has changed content.
Inputs: files (required). Returns: code, engine, status, checkId, +16 more
check_projectARead-onlyIdempotent
Runs and clusters Oxlint and TypeScript (and optionally Biome) lint and type diagnostics for one or more project paths. Read-only; no files are written or modified. Paths default to the project root (".") when omitted; paths must be relative and within the project directory — absolute paths or paths outside the root return an error response. Use this for a full project scan; use check_files instead for faster incremental checks after editing specific files. Each call re-runs all enabled engines with no caching.
Inputs: paths. Returns: code, engine, status, checkId, +16 more
compare_diagnosticsARead-onlyIdempotent
Computes a deterministic diagnostic delta between two checks (baselineCheckId and currentCheckId). Read-only; no files are modified. Identifies introduced errors, resolved errors, unchanged issues, and net error delta, resilient to line shifts.
Inputs: currentCheckId (required), baselineCheckId (required). Returns: code, status, message, netDelta, +10 more
discard_diagnostic_fixBIdempotent
Discards an in-memory diagnostic fix preview by transactionId, dropping it from server storage.
Inputs: transactionId (required). Returns: discarded, transactionId
get_diagnostic_snapshotARead-onlyIdempotent
Retrieves the immutable diagnostic snapshot recorded for a specific checkId. Read-only; no files are modified. Returns full metadata, engine statuses, clusters, remaining issues, and performance metrics for that check run.
Inputs: checkId (required). Returns: code, cache, source, status, +11 more
get_issue_detailARead-onlyIdempotent
Returns the full issue list for either one cluster ID or one issue ID for a specific checkId. Read-only; no files are written or modified. Supply checkId (required) and exactly one of clusterId or issueId — supplying both or neither returns an argument error. If the referenced checkId is unknown/expired or the issue/cluster no longer exists in that check snapshot, returns a status: "stale" response instead of an error.
Inputs: checkId (required), issueId, clusterId. Returns: code, issues, status, message, +1 more
get_live_diagnosticsBRead-onlyIdempotent
Returns current bounded active diagnostics and clusters merged from both internal engines and ingested LSP sources, with clear provenance per issue.
Inputs: files, severity. Returns: issues, status, clusters, totalIssues, +1 more
get_loop_statusARead-onlyIdempotent
Returns all diagnostic issue signatures currently flagged as looping (repeatedly appearing and disappearing) in this server session. Read-only; no files are written or modified. Loop history is accumulated across all check_project and check_files calls in this process lifetime, and is restored from .signalint/session.jsonl on startup. Takes no parameters. Use this to identify which diagnostics an agent is oscillating on; use check_project or check_files to run fresh diagnostics.
Returns: looping, signatures, fileChurning, fileRuleChurns
ingest_diagnosticsCRead-onlyIdempotent
Ingests external LSP diagnostics (e.g., from VS Code or language servers), normalizes them to Signalint's unified schema, updates the active diagnostic buffer, and records an immutable snapshot.
Inputs: diagnostics (required), source, serverName. Returns: checkId, clusters, snapshotId, totalIssues, +1 more
pingARead-onlyIdempotent
Checks whether the Signalint MCP server is responsive. Read-only; returns the string "pong" with no side effects. Use this to verify the server is connected before running diagnostics. Invalid arguments return an error response; no authentication is required.
Returns: pong, projectRoot
preview_diagnostic_fixARead-only
Creates an in-memory preview of proposed diagnostic fixes without touching the filesystem. Validates paths against canonical project root containment and returns a preview with a unique transactionId.
Inputs: patches (required). Returns: error, status, summary, filesCount, +2 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 13 tools
Most tools have clearly distinct purposes (check_project vs check_files vs check_project are distinguished by scope and caching; fix lifecycle tools are separated). Three tools — after_edit_check, compare_diagnostics, and ingest_diagnostics — all compute deltas or ingest diagnostics and could be confused in an agent's workflow, but descriptions clarify usage contexts well. Overall boundaries are clear enough that misselection is unlikely.
Names predominantly follow verb_noun (check_project, get_issue_detail, preview_diagnostic_fix, apply_diagnostic_fix, discard_diagnostic_fix, get_live_diagnostics), with a few noun_noun or truncated forms like after_edit_check, ping, and get_loop_status. The pattern is still very readable and mostly predictable, with minor deviations like ping (single verb) and snapshot retrieval names.
13 tools is a reasonable size for a diagnostics server that supports checking, snapshotting, comparing, previewing fixes, applying fixes, and ingesting LSP data. No tool feels redundant or trivially unnecessary, though the fix preview/apply/discard trio adds complexity that could be consolidated if needed.
The surface covers the full diagnostic lifecycle: connectivity (ping), full/incremental checking (check_project, check_files, after_edit_check), snapshot retrieval, delta comparison, live merged diagnostics, loop detection, fix preview/apply/discard, and external LSP ingestion. CRUD-like coverage for fixes and diagnostics is complete with no obvious dead ends.
Maintenance
Related MCP Connectors
The official Svelte MCP server providing docs and autofixing tools for Svelte development
Lean 4 MCP server: compile, prove theorems, and formalize math with Mathlib.
A TypeScript MCP server for Home Assistant, enabling programmatic management of entities, automati…
MCP server for understanding Javascript internals from ECMAScript specification.
Related MCP Servers
- AlicenseBqualityDmaintenanceAn MCP server that integrates the high-performance Oxlint linter into AI-powered editors and development tools. It enables efficient JavaScript and TypeScript code analysis and linting through the Model Context Protocol.110 npm1Apache 2.0
- AlicenseAqualityDmaintenanceA TypeScript-aware MCP server that provides coding agents with repository discovery, code intelligence, and web project context for local codebases. It enables deep symbol navigation, diagnostic reporting, and structural analysis of monorepos without requiring full IDE integration.711 npm1MIT
- AlicenseAqualityDmaintenanceA lightweight MCP server that provides 40 tools for TypeScript/JavaScript refactoring and code intelligence, directly mapping to TypeScript's tsserver protocol commands for accurate structural changes and workspace analysis.4029 npm3MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that provides TypeScript 7 native language server capabilities (go to definition, find references, hover types, diagnostics) to coding agents, using the Go-based tsc compiler for fast and accurate semantic analysis.2MIT