M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
M

MCPJam Inspector

Community
Not a standard server
Open-source testing and evaluations platform for MCP server developers

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #147 of 438
Stars
★ 2.3k Very popular
Transport
stdio (local process) · Streamable HTTP
Runtime
Node.js 20+ · Docker
Credentials
Optional API key
License
Other / unspecified
Last commit
58FMRS · C

MCPJam Inspector is a mature open-source MCP testing and debugging platform covering tool invocation, OAuth flows, cross-client evals, and CI/CD integration; the README is well documented with multiple deployment options, making it a solid debugging and quality-assurance tool for MCP server developers.

Strongest · Setup experience 13/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
13/20
Maintenance
11/20
Documentation
13/20
Setup experience
13/20
Why each score
Reliability 8/20
This is a static-only review; no server manifest was cached, and no committed CI workflows or test files were supplied as evidence. The README is largely marketing copy and offers no verifiable proof of a real MCP init/tool-list handshake or that declared capabilities match actual behavior. Notably, this project is better characterized as an inspector/client platform for testing other MCP servers rather than a conventional MCP server exposing tools — the README's sweeping claims (16 client configs, 170+ models) are unverified by any automated test or CI evidence in the provided material. Per the static calibration rule, reliability cannot exceed 12 without real CI plus committed tests covering key paths, so 8 is given, reflecting a plausible happy path with unverified edge-case/error handling.
Security and permissions 13/20
The evidence shows some security-conscious design: an OAuth debugger for diagnosing authorization flows; a stated guarantee that local skills read from the filesystem 'never leave your machine'; explicit Docker instructions to bind to 127.0.0.1 (not 0.0.0.0) with a stated security rationale; and the hosted app deliberately excluding STDIO/local file access to limit risk. No red-line issues (malware, hardcoded secrets, fabricated identity, unconfirmed destructive defaults) were found. However, the README does not detail how API keys, chat transcripts, or workspace data are handled/retained in the hosted service, nor does it describe confirmation steps for higher-risk operations (OAuth registration, outbound network calls), so permission scoping and data-flow disclosure remain incomplete.
Maintenance 11/20
The repository is not archived, and its star count (2136) and open-issue count (206) indicate meaningful activity and attention, but no close rate, release cadence, or version history was provided as evidence. More significantly, there is a licensing contradiction: the repo metadata reports 'NOASSERTION' (no asserted license), while the README badge and text claim 'Apache License 2.0' and point to a LICENSE file — this mismatch between stated and actual license metadata is treated as a deduction in a static review rather than resolved in the README's favor. No formal security-response channel is evident beyond a general Discord community link.
Documentation 13/20
The README is well organized, with layered sections (feature tables, install paths, Docker notes, contribution guide) and links out to docs.mcpjam.com for deeper documentation. Those external docs cannot be verified for accuracy or completeness within this static review. No in-repo tool parameter listing was available (manifest not cached), no troubleshooting section is present, and cost/rate-limit details for the 'free' chat offering are unspecified — leaving usable but incompletely disclosed documentation.
Setup experience 13/20
The README provides several clear install paths (hosted app, desktop installer, npx terminal run, Docker build-from-source), states the Node.js version requirement, and calls out a Docker security binding caveat — overall a well-documented setup story. Per the static calibration rule, setup is capped at 15 absent real CI workflows and committed tests demonstrating execution; since no such verifiable execution evidence was supplied, a score of 13 is given, reflecting clearly described but unverified setup steps.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Developers building or maintaining MCP servers who need to verify consistent behavior across many clients
  • Teams wanting to add MCP conformance and regression testing to their CI/CD pipeline
  • Developers debugging OAuth/DCR/CIMD authorization flows
  • Developers previewing and iterating on Apps SDK / MCP app widget UIs locally

Not for

  • End users who just want to plug an MCP server into an assistant client (this is a developer debugging tool, not an end-user MCP server)
  • Users needing hosted-only features like skill sync or tasks in a fully offline environment, since the hosted app is HTTPS-only and some features are local-only
  • Users looking for an MCP server implementation that exposes tools to an AI assistant

Required permissions

  • Network access to connect to target MCP servers (local STDIO process or remote HTTP/S endpoint)
  • The local/desktop build launches and communicates with local STDIO subprocesses, i.e. it can execute local commands
  • The OAuth debugger accesses and handles authorization credentials and token exchanges
  • A self-built Docker image binds to a local port (127.0.0.1 by default) requiring local port access

Risks and side effects

  • As a debugging tool it logs and displays full JSON-RPC messages and OAuth exchange details, so logs may contain sensitive tokens or credentials
  • Docker deployments that use `-p 6274:6274` instead of `-p 127.0.0.1:6274:6274` could expose the debugging port publicly
  • The hosted web app cannot be used for STDIO-only servers since it only supports HTTPS endpoints
  • The repository's license field is listed as NOASSERTION even though the README states Apache 2.0 — verify actual license terms before relying on this project

Setup

Before you start

Runtime：Node.js 20+ · Docker

Choose the installation method that fits your use case: 1) Hosted web app — open app.mcpjam.com, no install required, HTTPS server URLs only, no STDIO, tunneling, skills, or tasks; 2) Desktop app — download the Mac (.dmg) or Windows (.exe) installer from GitHub Releases, supports HTTP/S and local STDIO servers, no Node.js required; 3) Terminal — requires Node.js 20+, run npx @mcpjam/inspector@latest and open the printed localhost URL, supports HTTP/S and local STDIO; 4) Docker — no published image exists, so clone the repo and build from the included Dockerfile (docker build -t mcpjam/mcp-inspector:local -f mcpjam-inspector/Dockerfile .), always run with -p 127.0.0.1:6274:6274 (not -p 6274:6274) to keep it local-only, and on macOS/Windows reach host MCP servers via http://host.docker.internal:PORT.

Check that it works

Run npx @mcpjam/inspector@latest, open the printed localhost URL, and connect your MCP server — seeing its tools, resources, and prompts listed confirms the install works.

Troubleshooting

  1. Confirm the install method supports the transport you need: the hosted app is HTTPS-only, while STDIO servers require the desktop app or terminal `npx` install
  2. The terminal install requires Node.js 20+; an older Node version will cause the npx command to fail
  3. For Docker, verify the port mapping uses `127.0.0.1:6274:6274`, and on macOS/Windows use `host.docker.internal` instead of `localhost` to reach host-run MCP servers
  4. If OAuth debugging errors occur, check which protocol version (2025-03-26, 2025-06-18, 2025-11-25, or the 2026-07-28 draft) the target server implements against the versions the debugger supports
  5. If features like Skills, tasks, or tunneling appear missing, check whether you're using the hosted web app, since those are local/desktop-only features

Things to try

Once connected, you can ask your AI assistant things like:

  • Launch MCPJam Inspector with npx and open the localhost page to inspect my MCP server's tools
  • Use the Playground to test my server's tool calls with an LLM and view JSON-RPC traces
  • Run an OAuth conformance check against my MCP server
  • Run cross-client evals in CI to catch regressions

Use cases

Manually invoking and debugging an MCP server's tools, resources, and prompts with full JSON-RPC visibility during development
Debugging and validating MCP OAuth / EMA authorization flows and tracing errors across protocol versions
Running eval test cases across multiple LLM client configurations and tracking tool-call accuracy over time
Automating conformance, end-to-end tests, and evals in CI/CD to catch regressions before release
Iterating locally on widgets for servers that use the OpenAI Apps SDK or MCP app UIs

Overview

MCPJam Inspector is an open-source testing and evaluations platform for interactively inspecting and debugging MCP (Model Context Protocol) servers, MCP apps, and ChatGPT apps. It is not itself an MCP server exposing tools — it is an MCP client/debugging tool that connects to local (STDIO) or remote (HTTP/S) MCP servers, letting developers manually run tools, resources, resource templates, prompts, and elicitation flows while viewing full JSON-RPC logs. It supports cross-client evals across 16 client configurations (ChatGPT, Claude, Cursor, Copilot, and more) and 170+ models, includes a guided OAuth debugger covering multiple protocol versions (2025-03-26, 2025-06-18, 2025-11-25, and the 2026-07-28 draft, with DCR, client pre-registration, and CIMD support), a traced Playground chat interface with a Chrome DevTools-style widget emulator for OpenAI Apps SDK and MCP app UIs, Skills, team Workspaces, test-case Evals, and a CLI/SDK for CI/CD pipelines. It can be used via the hosted web app (app.mcpjam.com), a Mac/Windows desktop app, or run locally via npx @mcpjam/inspector@latest; a Docker image can also be built from source.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 608cb9d2884f Data synced 2026-10-19 Read the FMRS scoring method