bash-mcp
Provides ArgoCD application health and sync-status summaries to check whether apps are healthy and in sync.
Provides structured Git repository context and health summaries, including status, diffs, logs, and feature-branch review context.
Provides structured GitHub pull request CI check status and related PR context via the gh CLI.
Provides summaries of failed GitHub Actions workflow runs to help diagnose CI failures.
Provides structured GitLab CI pipeline and merge request job status, plus failure summaries for failed GitLab pipelines.
Provides Helm release triage tools to inspect release status and values and diagnose unhealthy Helm releases.
Provides structured Kubernetes diagnostics via kubectl, including pod failure diagnosis, namespace pod-failure summaries, deployment rollout health, and cluster event summaries.
Provides Liquibase tools to validate changelogs, preview SQL for pending changesets, and list changesets not yet applied to a database.
Provides structured summaries of Terraform plans so agents can review planned infrastructure changes without parsing raw plan output.
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., "@bash-mcp list all pods in the kube-system namespace with their restart counts"
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.
bash-mcp
An MCP server that wraps common CLI tools (kubectl, terraform, helm, git, argocd, jq, yq, ripgrep, etc.) and returns structured JSON instead of raw text output.
Designed for AI agents and LLM-powered workflows where parsing human-readable CLI output wastes tokens and breaks easily.
Created with the help of Claude since it is the main consumer of the structured data.
Quick Start
Option 1: Zero-install with npx
No installation needed — just add the MCP config and npx will fetch it on first use.
Option 2: Global install
npm install -g @nickw8/bash-mcpOption 3: Install from source
git clone https://github.com/nickw8/bash-mcp.git
cd bash-mcp
npm install && npm run build
npm linkRelated MCP server: Pare
Verify your setup
Run the built-in preflight check before wiring bash-mcp into a client:
npx -y @nickw8/bash-mcp --doctor # or: bash-mcp --doctor (global install)It reports your Node version, whether the MCP SDK loads, which CLIs are
available (versions included), and the resolved BASH_MCP_MODE — then exits
non-zero if a critical check fails (Node too old or the SDK can't load). Missing
CLIs are advisory, not failures. Without --doctor, the binary starts the MCP
server as usual.
Claude Code Setup
Add bash-mcp to your Claude Code settings. For all projects (recommended), edit ~/.claude/settings.json. For a single project, edit .claude/settings.json in the project root.
With npx (no install required):
{
"mcpServers": {
"bash-mcp": {
"command": "npx",
"args": ["-y", "@nickw8/bash-mcp"],
"env": {
"BASH_MCP_MODE": "readOnly"
}
}
}
}Safe by default: an unset
BASH_MCP_MODEresolves toreadOnly, which blocks mutatingrun/batchcommands. The examples set it explicitly for clarity. For trusted local use, setBASH_MCP_MODE=offto allow writes.
With global install:
{
"mcpServers": {
"bash-mcp": {
"command": "bash-mcp",
"env": {
"BASH_MCP_MODE": "readOnly"
}
}
}
}To verify it's working, start Claude Code and check that bash-mcp tools (like cat, rg, git_status) appear in the tool list.
You can also add it from the CLI:
claude mcp add bash-mcp -e BASH_MCP_MODE=readOnly -- npx -y @nickw8/bash-mcpSteer agents toward the structured tools (optional)
Registering the server exposes the tools, but agents still reach for raw Bash
out of habit. Two assets fix that: a rules file (~/.claude/rules/bash-mcp-tools.md,
auto-loaded into every session — the tool inventory) and a PreToolUse(Bash)
redirect hook. Install both with:
npx -y @nickw8/bash-mcp --install-claude # copy into ~/.claude
npx -y @nickw8/bash-mcp --install-claude --check # report drift, write nothingIt prints the settings.json hook block to paste in (the rules file needs no
settings — Claude Code auto-loads ~/.claude/rules/*.md). See
hooks/README.md for details and the clone-time
npm run claude:install equivalent. --uninstall-claude removes both files again.
Or install the Claude Code plugin, which bundles the server and the hook (in warn
mode — it never blocks unless BASH_MCP_HOOK=block):
/plugin install bash-mcp --marketplace nickw8/bash-mcpEvery other client (VS Code/Copilot, Codex, generic stdio) and pinned versions: docs/guides/installation.md.
Claude Desktop Setup
Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/,
Windows: %APPDATA%\Claude\) and add the same mcpServers entry:
{
"mcpServers": {
"bash-mcp": {
"command": "npx",
"args": ["-y", "@nickw8/bash-mcp"],
"env": {
"BASH_MCP_MODE": "readOnly"
}
}
}
}Restart Claude Desktop after editing.
Cursor Setup
Add bash-mcp to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):
{
"mcpServers": {
"bash-mcp": {
"command": "npx",
"args": ["-y", "@nickw8/bash-mcp"],
"env": {
"BASH_MCP_MODE": "readOnly"
}
}
}
}(Use "command": "bash-mcp" with no args if you installed globally.) Any MCP-aware
client works the same way — point it at the bash-mcp binary or npx -y @nickw8/bash-mcp.
Recommended Hooks
Optional Claude Code hooks in hooks/ steer agents toward the structured tools by blocking raw shell commands (git status, kubectl get, cat, …) that have a bash-mcp equivalent, while letting write commands (git commit, kubectl apply) pass through. BASH_MCP_HOOK=warn makes every block advisory; off disables the hook.
See docs/runbooks/hooks-setup.md for setup.
Why Structured Output?
Raw CLI tools return text meant for humans. AI agents waste tokens parsing tables, aligning columns, and guessing field boundaries. bash-mcp runs the same CLI tools but returns structured JSON with typed schemas, so the agent gets exactly the data it needs.
# Raw kubectl output: 148 tokens to parse a table
NAME READY STATUS RESTARTS AGE
nginx-7c5b8d6c88-abc12 1/1 Running 0 3d
# bash-mcp output: typed, compact, ready to use
{ "name": "nginx-7c5b8d6c88-abc12", "status": "Running", "age": "3d", ... }See docs/guides/token-benchmarks.md for measured savings (e.g.
kubectl describe pod → kube_diagnose_pod is ~64% fewer tokens).
Philosophy: structured tools first, run is the escape hatch
Reach for a structured tool first. Every wrapper returns typed JSON the agent can use directly — no column-counting, no re-parsing on a format change.
Prefer a diagnostic over a chain of raw commands. Tools like
kube_diagnose_pod,helm_release_triage, andargo_app_health_summaryanswer an operational question in one call —{ status, likelyCauses, suggestedNextCommands, evidence }— instead of making the agent runget, thenlogs, then reason across both.run/batchare the escape hatch. For commands without a dedicated wrapper,runexecutes anything with smart truncation. Optional hooks actively steer agents back to the structured tool when a raw equivalent is invoked.
Read-only tools carry the MCP readOnlyHint annotation. run/batch can be gated with
BASH_MCP_MODE (see Configuration).
Which tool should I use?
Pick the structured tool that matches your intent instead of reaching for a raw command.
The list_guidance tool returns this same index as JSON (filter by intent or category),
so agents can self-select without the README.
I want to… | Use this | Instead of |
diagnose a crashing or failing Kubernetes pod |
|
|
find all not-ready or failing pods in a namespace |
|
|
check a deployment's rollout health |
|
|
understand recent cluster events |
|
|
summarize the changes in a Terraform plan |
|
|
review the current feature branch before a PR |
|
|
understand the overall state of a repository |
|
|
read a file's structure without loading the full body |
|
|
read one region of a large file |
|
|
keep a broad search from flooding the context |
|
|
search for exact code references or symbols |
|
|
check whether a pull request's CI checks pass |
|
|
find out why a CI / GitHub Actions run failed |
|
|
check whether a GitLab pipeline or merge request's jobs pass |
|
|
find out why a GitLab CI pipeline failed |
|
|
triage why a Helm release is unhealthy |
|
|
check whether an ArgoCD app is healthy and in sync |
|
|
validate a Liquibase changelog for errors |
|
|
preview the SQL Liquibase would run for pending changesets |
|
|
list Liquibase changesets not yet applied to the database |
|
|
lint a shell script for bugs and quoting issues |
|
|
run a bats or shell test suite |
|
|
discover which CLIs are installed before calling a tool |
|
|
Tools
📖 Full tool reference →
docs/tools.md— generated from the registered tools (inputs, outputs, and the raw command each one approximates). Regenerate withnpm run docs:tools.
Environment
Tool | Description |
| Report which CLIs are installed (and their versions) so you can pick the right tool before calling it. |
| Return an intent → preferred-tool index so you can pick the right bash-mcp tool for a goal. |
Filesystem
Tool | Description |
| List files in a directory. |
| Show directory structure as a compact tree. |
| Show disk usage for paths. |
| Find files by name pattern, type, or modification time. |
Search
Tool | Description |
| Search file contents with ripgrep. |
| Find files matching a glob pattern. |
File
Tool | Description |
| Read one or more files with line numbers and smart truncation. |
| Show the structural outline of a file — function/class names, top-level comments, imports. |
Git
Tool | Description |
| Structured git status: branch, staged/unstaged/untracked files. |
| Structured git log: commit hash, author, date, message. |
| Structured git diff: files changed with insertions/deletions counts. |
| List git branches. |
| One-call snapshot of a git working tree: branch, ahead/behind vs upstream, staged/unstaged/untracked counts, recent commits, and the uncommitted diffstat. |
| Collect the commits and file changes of a branch vs a base ref (for writing a PR description): commit list, changed files with status, and a diffstat over base...head. |
| Show git diff with structured patch content. |
GitHub
Tool | Description |
| Report a pull request's CI checks as counts per state plus the failing and pending checks. |
| Diagnose a GitHub Actions run in one call: status, the failed jobs and steps, and the ##[error] lines from the failed log as evidence. |
GitLab
Tool | Description |
| Report a GitLab pipeline's jobs as counts per status plus the failed, running and pending jobs. |
| Diagnose a GitLab pipeline in one call: status, the failed jobs and why GitLab says they failed, and the last script lines of each failed job's trace as evidence. |
Kubernetes
Tool | Description |
| Get Kubernetes resources as structured data. |
| Get pod logs. |
| List available kubectl contexts with current context marked. |
| Diagnose why a pod is unhealthy in one call. |
| List unhealthy pods in a namespace with their failure reason and evidence — one call instead of get + describe per pod. |
| Report a deployment's rollout health (ready/desired replicas, conditions) as a structured diagnosis. |
| Summarize Warning events in a namespace (grouped by reason, ordered by count) instead of scrolling raw kubectl get events. |
Terraform
Tool | Description |
| List resources in Terraform state. |
| Show current Terraform state as structured JSON. |
| Run terraform plan and return a structured summary of changes (add/change/destroy counts and affected resources). |
| List Terraform workspaces with current workspace marked. |
| List Terraform/OpenTofu outputs as [name, value] pairs; a sensitive output is redacted to [name] alone. |
| List the Terraform/OpenTofu version and selected provider versions for the project. |
| Validate the Terraform/OpenTofu config and return a compact pass/fail summary with diagnostics. |
| List the modules used by an initialized Terraform/OpenTofu project (key, source, version). |
| Report the configured backend type and config for an initialized project. |
Helm
Tool | Description |
| List Helm releases with status, chart, and app version. |
| Get detailed status of a Helm release. |
| Get the computed values for a Helm release as structured data. |
| Diagnose a Helm release's health in one call: combines helm status + helm history into current status, likely causes, suggested next commands, and recent-revision evidence. |
ArgoCD
Tool | Description |
| List ArgoCD applications with sync/health status. |
| Get detailed status for a single ArgoCD application including resource health. |
| Show what's out of sync for an ArgoCD application. |
| Diagnose an ArgoCD application's health in one call: overall sync/health, likely causes, suggested next commands, and the unhealthy resources/conditions as evidence. |
Data Processing
Tool | Description |
| Query and transform JSON using jq expressions. |
| Query and transform YAML files using yq expressions (mikefarah/yq). |
.NET
Tool | Description |
| Run dotnet build and return structured diagnostics with file, line, column, message, and error code. |
| Run dotnet test and return structured results: pass/fail/skip counts, failure messages. |
Liquibase
Tool | Description |
| Validate a Liquibase changelog and return a structured pass/fail result with per-changeset errors (duplicate ids, checksum drift). |
| Render the SQL Liquibase would run for pending changesets (updateSQL) as structured per-changeset summaries with a SQL-Server batch lint. |
| List Liquibase changesets not yet applied to the target database, as structured JSON. |
Node.js
Tool | Description |
| Run biome check and return structured diagnostics with file, line, column, message, and rule. |
| Run vitest or jest (picked from package.json) and return structured test results: suites, pass/fail counts, failure messages. |
| Run tsc/tsgo --noEmit and return structured type errors with file, line, column, message, and TS error code. |
Python
Tool | Description |
| Run ruff check and return structured diagnostics with file, line, column, message, and rule code. |
| Run pytest and return structured test results: suites, pass/fail counts, failure messages. |
| Run mypy and return structured type errors with file, line, column, message, and error code. |
Shell
Tool | Description |
| Check shell scripts for syntax errors with |
| Run shellcheck and return structured diagnostics with file, line, column, message, and SC rule code. |
| Run a shell test script — bats |
Execution
Tool | Description |
| Run a single command — a binary plus an args array, executed directly without a shell (so no pipes, redirects, or ';'/'&&' in the string) — and return structured output with smart truncation. |
| Run an ordered list of labeled commands one after another, stopping at the first failure by default. |
| Run multiple commands in parallel and return all results. |
Results
Tool | Description |
| Page through a result the response budget trimmed. |
All tf_* tools accept binary: "terraform" | "tofu" (or set $TF_BINARY) to run OpenTofu.
Example Agent Workflows
Triage a crash-looping pod. Before — three raw calls plus reasoning:
kubectl get pods -n payments # find the broken pod
kubectl describe pod api-xyz # read events/state, infer cause
kubectl logs api-xyz --tail=200 # scroll for the error
# ...agent reasons across all three to guess the next stepAfter — one structured call:
kube_diagnose_pod(pod="api-xyz", namespace="payments")
→ { status: "CrashLoopBackOff",
likelyCauses: ["container exits 1; DB connection refused"],
suggestedNextCommands: ["kube_logs(pod='api-xyz', container='api')"],
evidence: ["Restart Count: 12", "BackOff x140 over 30m"] }Review a feature branch. Before: git log main..HEAD, git diff --stat main...HEAD,
git diff main...HEAD — three commands, manual stitching. After:
git_pr_context(base="main")
→ { commits: [...], files: [...], summary: { filesChanged, insertions, deletions } }Check capabilities before acting. Instead of which kubectl && kubectl config current-context:
check_environment()
→ { kubectl: { installed: true, version: "1.31.2", context: "prod" },
terraform: { installed: false }, ... }Configuration
Environment Variables
Variable | Values | Default | Effect |
|
|
| Wide-event JSON logging to stderr (one line per tool call). |
|
|
| Gates |
|
|
| Which tool groups register. Fewer tools, less schema in context — see tool discovery. An unknown value fails at startup. |
|
|
| Response budget. A cut result carries |
| seconds |
| How long a stored result stays retrievable; |
|
|
| Redirect hook behaviour. |
|
|
| Default binary for all |
Timeouts
Each tool group uses appropriate default timeouts:
Filesystem/search/git: 30 seconds
Kubernetes/Helm/ArgoCD: 15 seconds
Terraform state: 30 seconds
Terraform plan: 120 seconds
.NET build/test: 120 seconds
Python lint: 30 seconds
Python test: 120 seconds
Python typecheck: 60 seconds
Command Buffer
All commands default to a 10 MB output buffer. Override per-tool in src/exec.ts.
Development
# Install dependencies
npm install
# Run in development mode (auto-reloads)
npm run dev
# Build for production
npm run build
# Start production server
npm start
# Run tests
npm test
# Lint and format
npm run lint
npm run formatArchitecture
src/
index.ts # Server entry point — registers tools and starts stdio transport
exec.ts # Command execution layer — runs CLI tools via execFile
tool.ts # defineTool — wraps registerTool with wide-event logging + error catching
error.ts # ToolError taxonomy + classifyError (missing_binary, timeout, ...)
logger.ts # Zero-dep structured stderr logger (BASH_MCP_LOG)
safety.ts # Command classification + BASH_MCP_MODE gate for run/batch
format.ts # Multi-format list output (TSV, columnar, JSON)
response.ts # MCP response helpers (ok, okList, err)
shell.ts # Shell escaping utilities
parsers/
types.ts # Shared interfaces: Diagnostic, TestResult, TestSuite
schemas.ts # Shared Zod (diagnostic, budget fragment, applyBudget)
tools/
argocd/argocd.ts # argo_apps, argo_app_detail, argo_app_diff
batch/batch.ts # batch (parallel command execution)
dotnet/
dotnet.ts # dotnet_build, dotnet_test
parsers/ # MSBuild and TRX output parsers
file/
file.ts # cat (file reading with metadata), outline (structural file outline)
outline/ # Language-specific outline extractors
filesystem/
filesystem.ts # ls, tree, du, find_files
git/
git.ts # git_status, git_log, git_diff, git_branches
diff.ts # git_diff_content (structured patch hunks)
helm/helm.ts # helm_list, helm_status, helm_values
json/json.ts # jq
kubernetes/
kubernetes.ts # kube_get, kube_logs, kube_contexts
npm/
npm.ts # npm_lint, npm_test, npm_typecheck
parsers/ # Biome, Vitest, tsc output parsers
run/run.ts # run (command execution with smart truncation)
search/search.ts # rg, glob
terraform/
terraform.ts # tf_state_list, tf_show, tf_plan_summary, tf_workspaces
yaml/yaml.ts # yqEach tool file exports a single register*Tools(server) function that registers all tools in its group. Every tool returns both content (text for display) and structuredContent (typed JSON for programmatic use).
Adding a New Tool
Add to an existing category (
src/tools/<group>/<group>.ts) or create a new category directoryDefine input/output schemas with Zod (add
annotations: { readOnlyHint: true }for read-only tools)Implement the handler using
exec()orexecJson()from#exec, returnok()orerr()from#responseRegister it with
defineToolfrom#tool(injects wide-event logging + uniform error catching)Wire the group's
register*Toolsintosrc/index.tsAdd tests in a co-located
*.test.tsfile
import { z } from "zod";
import type { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { exec } from "#exec";
import { ok, err } from "#response";
import { defineTool } from "#tool";
export function registerMyTools(server: McpServer) {
defineTool(server, "my_tool", {
title: "My tool",
description: "What it does and why structured output helps.",
inputSchema: {
path: z.string().describe("Path to operate on"),
},
outputSchema: {
result: z.string(),
},
annotations: { readOnlyHint: true },
}, async ({ path }) => {
const result = await exec("mytool", ["--json", path]);
if (result.exitCode !== 0) {
return err(result.stderr, { result: "" });
}
return ok({ result: result.stdout.trim() });
});
}See docs/guides/adding-tools.md for the full guide.
Prerequisites
The following CLI tools must be installed for their respective tool groups to work:
Always required: Node.js >= 20
Filesystem/search:
ls,tree,find,du,rg(ripgrep)Git:
gitKubernetes:
kubectlTerraform/OpenTofu:
terraformortofuHelm:
helmArgoCD:
argocdJSON:
jqYAML:
yq(mikefarah/yq).NET:
dotnet(.NET SDK)Liquibase:
liquibasePython:
ruff(lint),mypy(typecheck),pytest(test)
Tools gracefully return errors if their underlying CLI is not installed.
License
Apache-2.0
Available Tools
65 toolsargo_app_detailArgoCD app detailARead-only
Get detailed status for a single ArgoCD application including resource health.
Inputs: name (required). Returns: name, message, project, retained, +5 more
argo_app_diffArgoCD app diffBRead-only
Show what's out of sync for an ArgoCD application.
Inputs: name (required). Returns: diff, hasDiff, retained
argo_app_health_summaryArgoCD app health summaryARead-only
Diagnose an ArgoCD application's health in one call: overall sync/health, likely causes, suggested next commands, and the unhealthy resources/conditions as evidence.
Inputs: name (required). Returns: name, status, healthy, evidence, +4 more
argo_appsArgoCD applicationsBRead-only
List ArgoCD applications with sync/health status. Structured summary instead of table output.
Inputs: project, selector. Returns: apps, count, summary, retained
bash_lintBash lint (structured)ARead-only
Run shellcheck and return structured diagnostics with file, line, column, message, and SC rule code. Much more compact than raw shellcheck output. Use minSeverity to filter by severity level.
Inputs: files, maxItems, detailLevel, minSeverity. Returns: total, errors, retained, truncated, +2 more
bash_syntax_checkBash syntax check (structured)ARead-only
Check shell scripts for syntax errors with bash -n and return structured diagnostics (file, line, message). Parses without executing, so it is safe on untrusted scripts. Reports valid:true when no errors are found.
Inputs: files, maxItems, detailLevel. Returns: total, valid, errors, retained, +2 more
bash_testBash test (structured)A
Run a shell test script — bats .bats files via --tap, or a plain .sh harness — and return structured results: per-case pass/fail plus a summary parsed from TAP or N tests, M failures output, falling back to exit-code-only when the format is unrecognized. Executes the script, so it is gated by BASH_MCP_MODE.
Inputs: cwd, file, timeout. Returns: failed, passed, skipped, summary, +2 more
batchRun commands in parallelA
Run multiple commands in parallel and return all results. Each command is a binary plus an args array, executed without a shell (for a pipeline use command='sh', args=['-c', '...']). Use when you need to gather independent pieces of information in a single tool call instead of multiple sequential calls.
Inputs: commands. Returns: elapsed, results, retained
catRead file contentsARead-only
Read one or more files with line numbers and smart truncation. Returns structured output with metadata. Pass path for a single file, or paths to read several in one call (returns { files, count }) — collapsing what would otherwise be multiple round-trips. For large files, use startLine/endLine or maxLines to limit output. Use ref to read from a git branch/commit instead of disk (e.g. ref='main'). NOTE: reading with cat does NOT satisfy the built-in Edit/Write "must read first" guard (it tracks only the built-in Read tool) — run the built-in Read on a file immediately before editing it.
Inputs: ref, path, paths, endLine, +3 more. Returns: path, size, count, error, +7 more
check_environmentCheck environmentARead-only
Report which CLIs are installed (and their versions) so you can pick the right tool before calling it. Probes are client-only and non-blocking — no server/auth calls. Includes kubectl's current context when available.
Inputs: no named parameters. Returns: tools, total, retained, installedCount
dotnet_buildBuild (structured)A
Run dotnet build and return structured diagnostics with file, line, column, message, and error code. Much more compact than raw MSBuild output. Warnings are omitted by default; pass includeWarnings:true (or detailLevel:'full') to include them.
Inputs: cwd, project, maxItems, detailLevel, +2 more. Returns: total, summary, exitCode, retained, +4 more
dotnet_testTest (structured)A
Run dotnet test and return structured results: pass/fail/skip counts, failure messages. Much more compact than raw test output. Only failures are listed.
Inputs: cwd, filter, project. Returns: total, failed, passed, skipped, +4 more
duDisk usageARead-only
Show disk usage for paths. Returns entries as [path, sizeBytes] pairs, in the order du emits them. Use maxDepth to control how deep the summary goes.
Inputs: path (required), maxDepth. Returns: entries, retained
find_filesFind filesARead-only
Find files by name pattern, type, or modification time. Returns structured list of paths. Use names for multiple glob patterns (OR logic), modifiedWithin for recent files (e.g. '7d', '1h').
Inputs: name, path, type, names, +4 more. Returns: count, files, metadata, retained
gh_pr_checksPR checksARead-only
Report a pull request's CI checks as counts per state plus the failing and pending checks. Defaults to the current branch's PR; passing checks are counted, not listed, unless all=true.
Inputs: pr, all, cwd. Returns: checks, counts, retained
gh_run_failure_summaryCI run failure summaryARead-only
Diagnose a GitHub Actions run in one call: status, the failed jobs and steps, and the ##[error] lines from the failed log as evidence. Defaults to the latest run on the current branch.
Inputs: cwd, run, branch. Returns: run, url, branch, status, +6 more
git_branchesGit branchesBRead-only
List git branches. Returns the current branch and [name, lastCommit] pairs; a remote branch's name starts with remotes/.
Inputs: cwd, remote. Returns: current, branches, retained
git_diffGit diffARead-only
Structured git diff: files changed with insertions/deletions counts. Use base+ref for two-ref comparison (e.g. base='main', ref='feature'). Use path to scope to one or more files or directories.
Inputs: cwd, ref, base, path, +1 more. Returns: files, retained, fileCount, totalDeletions, +1 more
git_diff_contentGit diff with contentARead-only
Show git diff with structured patch content. Returns parsed hunks per file instead of raw unified diff text. Use this when you need to see actual code changes, not just file-level stats. Use base+ref for two-ref comparison (e.g. base='main', ref='feature'). Use path to scope to one or more files or directories.
Inputs: cwd, ref, base, path, +2 more. Returns: files, summary, retained
git_logGit logARead-only
Structured git log: commit hash, author, date, message. Compact and parseable. Use range for ref comparisons (e.g. 'main..feature'). Use exclude to filter out commits matching a pattern (e.g. 'ci:').
Inputs: cwd, path, count, range, +4 more. Returns: count, commits, retained
git_pr_contextGit PR contextARead-only
Collect the commits and file changes of a branch vs a base ref (for writing a PR description): commit list, changed files with status, and a diffstat over base...head.
Inputs: cwd, base, head. Returns: base, head, files, changes, +2 more
git_statusGit statusARead-only
Structured git status: branch, staged/unstaged/untracked files. Replaces parsing raw git status output.
Inputs: cwd. Returns: ahead, clean, behind, branch, +4 more
glab_ci_statusGitLab pipeline statusARead-only
Report a GitLab pipeline's jobs as counts per status plus the failed, running and pending jobs. Defaults to the current branch's latest pipeline; pass mr for a merge request's head pipeline. Passing, skipped and manual jobs are counted, not listed, unless all=true.
Inputs: mr, all, cwd, repo, +2 more. Returns: url, jobs, counts, status, +2 more
glab_pipeline_failure_summaryGitLab pipeline failure summaryARead-only
Diagnose a GitLab pipeline in one call: status, the failed jobs and why GitLab says they failed, and the last script lines of each failed job's trace as evidence. Allowed-to-fail jobs are noted, not counted as the failure. Defaults to the current branch's latest pipeline.
Inputs: mr, cwd, repo, branch, +1 more. Returns: url, branch, status, evidence, +5 more
globGlob file searchBRead-only
Find files matching a glob pattern. Returns a compact list of paths.
Inputs: pattern (required), cwd. Returns: count, files, retained
helm_listHelm releasesBRead-only
List Helm releases with status, chart, and app version. Structured output.
Inputs: filter, context, namespace, allNamespaces. Returns: count, releases, retained
helm_release_triageHelm release triageARead-only
Diagnose a Helm release's health in one call: combines helm status + helm history into current status, likely causes, suggested next commands, and recent-revision evidence.
Inputs: release (required), context, namespace. Returns: status, healthy, evidence, retained, +4 more
helm_statusHelm release statusCRead-only
Get detailed status of a Helm release.
Inputs: release (required), context, namespace. Returns: name, notes, status, retained, +4 more
helm_valuesHelm release valuesBRead-only
Get the computed values for a Helm release as structured data.
Inputs: release (required), context, allValues, namespace. Returns: values, retained
jqjq JSON processorARead-only
Query and transform JSON using jq expressions. Accepts a file path or raw JSON string. Returns parsed, structured output — far more compact than reading a full JSON file. The filter is passed directly to jq (no shell), so shell-style escaping such as backslash-escaped quotes is neither needed nor supported — write the filter exactly as jq expects it.
Inputs: file, input, slurp, filter, +2 more. Returns: result, retained, multiline
kube_contextsKubectl contextsARead-only
List available kubectl contexts with current context marked.
Inputs: no named parameters. Returns: current, contexts, retained
kube_deployment_statusDeployment rollout statusBRead-only
Report a deployment's rollout health (ready/desired replicas, conditions) as a structured diagnosis.
Inputs: name (required), context, namespace. Returns: status, evidence, retained, likelyCauses, +1 more
kube_diagnose_podDiagnose a podARead-only
Diagnose why a pod is unhealthy in one call. Returns status, likely causes, suggested next commands, and evidence (CrashLoopBackOff/ImagePullBackOff/OOMKilled/Unschedulable/restarts).
Inputs: pod (required), context, namespace. Returns: status, evidence, retained, likelyCauses, +1 more
kube_events_summarySummarize warning eventsARead-only
Summarize Warning events in a namespace (grouped by reason, ordered by count) instead of scrolling raw kubectl get events.
Inputs: context, maxItems, namespace, includeRaw, +2 more. Returns: total, status, evidence, retained, +3 more
kube_getKubectl get resourcesARead-only
Get Kubernetes resources as structured data. Wraps kubectl get -o json and returns a compact summary by default. Use the jq param to extract specific fields from the raw JSON (e.g. '.spec.template.spec.containers[].env') instead of the summary.
Inputs: jq, name, context, maxItems, +7 more.
kube_logsKubectl logsARead-only
Get pod logs. lines are as kubectl prints them, ' ', oldest first; consecutive repeats of a message merge into the first, marked ' [×N]', and a budget cut keeps the newest. Use grep to filter lines by regex pattern (e.g. 'ERROR|WARN') instead of piping through shell commands.
Inputs: pod, grep, tail, since, +7 more. Returns: lines, total, retained, truncated
kube_pod_failure_summarySummarize failing podsARead-only
List unhealthy pods in a namespace with their failure reason and evidence — one call instead of get + describe per pod.
Inputs: context, maxItems, namespace, includeRaw, +2 more. Returns: pods, total, status, evidence, +4 more
liquibase_statusPending changesets (structured)ARead-only
List Liquibase changesets not yet applied to the target database, as structured JSON. Reports up-to-date vs a pending list. liquibase is on PATH — no mise/wrapper needed.
Inputs: cwd, labels, contexts, extraArgs, +2 more. Returns: pending, retained, upToDate, pendingCount
liquibase_update_sqlPreview migration SQL (structured)ARead-only
Render the SQL Liquibase would run for pending changesets (updateSQL) as structured per-changeset summaries with a SQL-Server batch lint. Does NOT apply changes. SQL bodies are omitted unless includeRaw or changesetId is set. liquibase is on PATH — no mise/wrapper needed.
Inputs: cwd, labels, contexts, maxItems, +7 more. Returns: total, retained, truncated, changesets, +1 more
liquibase_validateValidate changelog (structured)ARead-only
Validate a Liquibase changelog and return a structured pass/fail result with per-changeset errors (duplicate ids, checksum drift). Much more compact than raw output. liquibase is on PATH — no mise/wrapper needed.
Inputs: cwd, labels, contexts, extraArgs, +2 more. Returns: valid, errors, retained, errorCount
list_guidanceList tool-selection guidanceARead-only
Return an intent → preferred-tool index so you can pick the right bash-mcp tool for a goal. Each entry names the tool to prefer, raw-command anti-patterns to avoid, and why. Filter with intent (substring) or category to narrow the list.
Inputs: intent, category. Returns: total, intents, retained
lsList directoryARead-only
List files in a directory. Returns entries as [name, size, permissions, modified] — directories end with '/' and links with '@'. Use nameOnly for names alone, recursive for one level of subdirectories, all to include hidden files.
Inputs: all, path, nameOnly, recursive. Returns: entries, retained
npm_lintLint (structured)A
Run biome check and return structured diagnostics with file, line, column, message, and rule. Far more compact and parseable than raw lint output. Use minSeverity to filter by severity level.
Inputs: cwd, fix, paths, maxItems, +2 more. Returns: total, errors, retained, truncated, +3 more
npm_testTest (structured)A
Run vitest or jest (picked from package.json) and return structured test results: suites, pass/fail counts, failure messages. Much more compact than raw test output. Only failures are listed by default; use verbose=true for all tests.
Inputs: cwd, pattern, verbose, coverage. Returns: suites, summary, retained
npm_typecheckType check (structured)ARead-only
Run tsc/tsgo --noEmit and return structured type errors with file, line, column, message, and TS error code. Much more compact than raw tsc output. Auto-detects tsgo for faster checking. Use project to specify a tsconfig.
Inputs: cwd, project, maxItems, detailLevel. Returns: total, errors, success, retained, +2 more
outlineFile structure outlineARead-only
Show the structural outline of a file — function/class names, top-level comments, imports. Returns a compact view without implementation bodies. Includes git metadata (branch, commit) when the file is in a git repo. Use instead of cat when reviewing file structure or auditing many files at once.
Inputs: path (required), ref. Returns: path, mtime, branch, commit, +6 more
python_lintPython lint (structured)B
Run ruff check and return structured diagnostics with file, line, column, message, and rule code. Much more compact than raw ruff output. Use minSeverity to filter by severity level.
Inputs: cwd (required), fix, paths, maxItems, +2 more. Returns: total, errors, retained, truncated, +3 more
python_testPython test (structured)A
Run pytest and return structured test results: suites, pass/fail counts, failure messages. Much more compact than raw test output. Only failures are listed by default; use verbose=true for all tests.
Inputs: cwd (required), path, pattern, verbose. Returns: suites, summary, retained
python_typecheckPython type check (structured)ARead-only
Run mypy and return structured type errors with file, line, column, message, and error code. Much more compact than raw mypy output.
Inputs: cwd (required), paths, maxItems, detailLevel. Returns: total, errors, success, retained, +2 more
repo_health_summaryRepo health summaryARead-only
One-call snapshot of a git working tree: branch, ahead/behind vs upstream, staged/unstaged/untracked counts, recent commits, and the uncommitted diffstat.
Inputs: cwd, commits. Returns: ahead, clean, behind, branch, +6 more
result_getGet Retained ResultARead-only
Page through a result the response budget trimmed. Pass the retained.id from that response; path selects a field (dotted, e.g. files.0), offset/limit slice an array. Snapshot of the original call, kept for retained.expiresInSec.
Inputs: id, path, limit, offset. Returns: id, path, tool, total, +4 more
rgRipgrep searchARead-only
Search file contents with ripgrep. Returns files as [file, "line:text", ...] rows (a context line is "line-text"), [file, count] under countPerFile, [file] under filesOnly. Use glob (string or array, '!' to exclude) to filter files, ignoreCase for case-insensitive search, filesOnly for just filenames, countPerFile for match counts per file, maxPerFile to cap hits per file. For 'collect all X' tasks use only:true to return just the matched substrings (one row per hit) — add replace ($1 capture groups) to extract structured values.
Inputs: glob, only, path, context, +9 more. Returns: files, retained, truncated, matchCount, +1 more
runRun a commandA
Run a single command — a binary plus an args array, executed directly without a shell (so no pipes, redirects, or ';'/'&&' in the string) — and return structured output with smart truncation. Pass the binary as command and its arguments as args (e.g. command='npm', args=['test']); for a real shell pipeline run the shell yourself: command='sh', args=['-c', 'a | b > c']. Keeps the last N lines of output by default (where errors typically appear), or the first N with mode='head'. Use for build, test, and lint commands where you need exit code and error details, not full verbose output.
Inputs: command (required), cwd, args, mode, +4 more. Returns: stderr, stdout, elapsed, exitCode, +3 more
run_seqRun commands in sequenceA
Run an ordered list of labeled commands one after another, stopping at the first failure by default. Use for dependent steps (e.g. build then test then package) where order matters and a later step is pointless if an earlier one fails. Unlike batch (parallel), run_seq is sequential and short-circuits; set stopOnError=false to run every step regardless.
Inputs: steps, maxLines, stopOnError. Returns: steps, elapsed, exitCode, failedAt, +1 more
tf_backend_infoTerraform backend infoARead-only
Report the configured backend type and config for an initialized project. Reads .terraform/terraform.tfstate — run init first.
Inputs: cwd (required). Returns: type, config, retained
tf_modules_summaryTerraform modules summaryARead-only
List the modules used by an initialized Terraform/OpenTofu project (key, source, version). Reads .terraform/modules/modules.json — run init first.
Inputs: cwd (required). Returns: count, modules, retained
tf_outputsTerraform outputsARead-only
List Terraform/OpenTofu outputs as [name, value] pairs; a sensitive output is redacted to [name] alone.
Inputs: cwd (required), binary. Returns: outputs, retained
tf_plan_summaryTerraform plan summaryARead-only
Run terraform plan and return a structured summary of changes (add/change/destroy counts and affected resources). Each action is create, update, delete, replace or read; a replace counts in add and destroy, as terraform's Plan line does.
Inputs: cwd (required), binary, target, varFile, +1 more. Returns: add, change, changes, destroy, +2 more
tf_providersTerraform providersCRead-only
List the Terraform/OpenTofu version and selected provider versions for the project.
Inputs: cwd (required), binary. Returns: version, retained, providers
tf_showTerraform showBRead-only
Show current Terraform state as structured JSON. Returns resource summary with types and attributes.
Inputs: cwd (required), binary. Returns: count, retained, resources
tf_state_listTerraform state listARead-only
List resources in Terraform state. Returns the resource addresses, plus a byType rollup once there are more than 20 (type, name, and module are readable from each address).
Inputs: cwd (required), binary. Returns: byType, retained, resources
tf_validate_summaryTerraform validate summaryARead-only
Validate the Terraform/OpenTofu config and return a compact pass/fail summary with diagnostics.
Inputs: cwd (required), binary. Returns: valid, retained, errorCount, diagnostics, +1 more
tf_workspacesTerraform workspacesARead-only
List Terraform workspaces with current workspace marked.
Inputs: cwd (required), binary. Returns: current, retained, workspaces
treeDirectory treeARead-only
Show directory structure as a compact tree. Returns structured nodes instead of ASCII art. Use dirsOnly for directory-only view, pattern to filter by glob, maxDepth to control depth.
Inputs: path (required), exclude, pattern, dirsOnly, +1 more. Returns: dirs, files, paths, retained
yqyq YAML processorARead-only
Query and transform YAML files using yq expressions (mikefarah/yq). Outputs as JSON for structured consumption. Far more compact than reading a full YAML file.
Inputs: file, input, expression, outputFormat. Returns: format, result, retained
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 65 tools
Most tools target a distinct tool/resource (kube vs helm vs argo vs tf vs per-language lint/test), but several pairs are near-neighbors: find_files vs glob, git_diff vs git_diff_content, tf_state_list vs tf_show, and the cluster of '*_health_summary'/'*_failure_summary'/'*_diagnose' triage tools overlap conceptually with the plain getters they wrap. Descriptions do resolve most of these, but at 65 tools an agent must read carefully to avoid picking e.g. kube_get over kube_diagnose_pod when a diagnosis is wanted.
Everything is snake_case, which is good, but the prefix/verb ordering is mixed: domain-prefixed noun-first (git_status, kube_get, tf_plan_summary, argo_app_diff) coexists with verb-first (list_guidance, check_environment, result_get) and bare Unix-style names (ls, tree, du, rg, glob, cat, jq, yq, run, batch). Readable, but not a single predictable pattern.
65 tools is very heavy, even for a broad shell/devops toolbox spanning git, k8s, Terraform, ArgoCD, Helm, four language toolchains, and CI. Each tool is defensible in isolation, but the surface invites selection errors and could be consolidated (e.g. per-language lint/typecheck/test trios, summary vs getter variants).
Read/inspect coverage is deep across every advertised domain (files, git, Terraform, k8s, ArgoCD, Helm, CI, language toolchains), and run/batch act as an escape hatch for anything missing. Gaps are mostly write-side: no git mutations (commit/checkout/push), no kube apply/scale/exec, no file edit—though run largely works around them.
Maintenance
Related MCP Connectors
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
Unified API to query AWS, GCP, Azure and generate Terraform/CLI execution kits for AI agents.
Deterministic JSON repair, validate, example-gen, schema-coerce for agents. Zero LLM, sub-10ms.
Convert, lint and generate LLM tool schemas for OpenAI, Anthropic, Gemini and MCP
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceProvides MCP servers that wrap common developer tools such as git, npm, Docker, Kubernetes, and test runners, returning schema-validated structured JSON instead of raw terminal text. This lets AI agents consume typed, token-efficient results directly without brittle string parsing or platform-dependent formatting issues.4 npmMIT
- AlicenseNot gradedqualityAmaintenanceWraps developer tools like git, npm, Docker, and test runners to return clean, schema-validated JSON instead of raw terminal text. This lets agents use typed, token-efficient results directly without brittle string parsing.4 npmMIT
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to work with developer tools such as git, npm, Docker, and test runners through structured, schema-validated JSON outputs. Reduces token usage and eliminates fragile terminal text parsing for reliable cross-platform workflows.4 npmMIT
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to invoke common developer tools such as git, npm, Docker, and test runners and receive clean, schema-validated JSON instead of fragile terminal text, reducing token usage and parsing errors.4 npmMIT