MCPJam MCP server
Commit Nov 14, 2026Lets AI clients chat with, inspect, and debug MCP servers, tools, prompts, and OAuth flows with full JSON-RPC traces.
Also known as MCPJam/inspector, @mcpjam/cli, @mcpjam/sdk
- Stars
- 2,318
- Tools
- 17
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 14, 2026
- License
- —
What is MCPJam MCP?
MCPJam lets AI clients chat with, inspect, and debug MCP servers, tools, prompts, and OAuth flows with full JSON-RPC traces. It exposes 17 tools, such as connect_server, disconnect_server, list_servers.
Summary generated from the project's README
About MCPJam MCP server
Testing and evaluating MCP servers is what inspector enables for developers who ship production-ready integrations. It provides a comprehensive platform to inspect tools, prompts, resources, and authorization flows with full visibility into JSON-RPC traces. Developers can use the hosted web app for HTTPS servers or run the tool locally via npx or a desktop installer to support local STDIO and HTTP/S connections.
The platform helps identify how different clients like Claude, ChatGPT, and Cursor interpret your server by offering cross-client evaluations across 16 configurations and 170 models. It includes a specialized OAuth debugger to visualize requests and check conformance against multiple protocol versions. Beyond manual debugging, the system supports automated testing through a CLI and SDK, allowing teams to integrate conformance and behavior checks directly into CI/CD pipelines like GitHub Actions.
What you can do with it
- 1Debug JSON-RPC communicationA developer can watch every tool call and agent step through detailed message logs.
- 2Verify OAuth conformanceYou can use the guided debugger to ensure your authorization flows meet specific protocol versions.
- 3Automate regression testingTeams can run automated checks on every pull request to prevent breaking changes.
- 4Evaluate model performanceRun test cases against various LLMs to track how accurately they use your tools over time.
Written from the project's README. Every use case is checked against a quote from it.
How to install MCPJam MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall MCPJam MCP in Claude Code
$ claude mcp add inspector -- npx -y @mcpjam/cli mcpInstall MCPJam MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"inspector": {
"command": "npx",
"args": [
"-y",
"@mcpjam/cli",
"mcp"
]
}
}
}Install MCPJam MCP in VS Code
.vscode/mcp.json
{
"servers": {
"inspector": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@mcpjam/cli",
"mcp"
]
}
}
}Install MCPJam MCP in VS Code CLI
$ code --add-mcp '{"name":"inspector","type":"stdio","command":"npx","args":["-y","@mcpjam/cli","mcp"]}'Install MCPJam MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"inspector": {
"command": "npx",
"args": [
"-y",
"@mcpjam/cli",
"mcp"
]
}
}
}Install MCPJam MCP in Codex
$ codex mcp add inspector -- npx -y @mcpjam/cli mcpInstall MCPJam MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.inspector] command = "npx" args = ["-y", "@mcpjam/cli", "mcp"]
Install MCPJam MCP in Gemini CLI
$ gemini mcp add -s user inspector npx -- -y @mcpjam/cli mcpInstall MCPJam MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"inspector": {
"command": "npx",
"args": [
"-y",
"@mcpjam/cli",
"mcp"
]
}
}
}Started by HostedMCP on 16 Nov 2026
We ran this command for @mcpjam/cli 6.3.1 in an isolated container with no network and no keys. It answered and listed 17 tools.
MCPJam MCP tools 17
Read from the server with tools/listconnect_serverActionOpen a persistent connection to an MCP server under test, over HTTP (url) or stdio (command). The connection stays open across tool calls so session state and notifications can be observed. Returns the negotiated initialization info.
disconnect_serverActionClose an open connection created by connect_server and discard its buffered notifications.
list_serversRead-onlyList the MCP server connections opened with connect_server, with status and a redacted target summary.
server_infoRead-onlyGet the negotiated initialization info for a connected server: protocol version, transport, server info/version, capabilities, and instructions.
ping_serverRead-onlySend an MCP ping over an open connection and report the round-trip latency.
list_toolsRead-onlyList tools exposed by a connected server, including pagination cursor and per-tool metadata.
call_toolActionInvoke a tool on a connected server and return the target's raw CallToolResult. A tool-level failure is reported inside the payload via its "isError" field; this call only errors when the request itself fails.
list_resourcesRead-onlyList resources exposed by a connected server. Set templates=true to list resource templates instead.
read_resourceRead-onlyRead a resource from a connected server by URI.
list_skillsRead-onlyList the Agent Skills a connected server serves over the skills extension (SEP-2640), including ones it advertises that cannot be verified, each with the reason.
get_skillRead-onlyFetch one skill by uri and verify it before returning content: the SKILL.md digest against the manifest, the fetched frontmatter against what the listing advertised, and the name against the uri. Answers with the skill, or with a refusal naming the check that failed.
read_skill_fileRead-onlyRead one supporting file of a skill served by a connected server, checked against that skill's own manifest for byte length and digest.
Show all 17 toolsShow fewer tools
list_promptsRead-onlyList prompts exposed by a connected server.
get_promptRead-onlyFetch a named prompt from a connected server, optionally with string arguments.
get_notificationsActionReturn notifications received from connected servers since connect (log messages, progress, tools/resources/prompts list_changed, resource updates). Use after call_tool or a server-side change to verify the target emitted the expected notifications.
server_doctorRead-onlyRun a stateless diagnostic sweep against an MCP server (HTTP url or stdio command) without keeping a connection open: reachability, initialize, capability checks, and tool/resource/prompt counts. Equivalent to `mcpjam server doctor`.
probe_serverRead-onlyProbe an HTTP MCP server without the full client connect flow: transport selection, auth requirements, and OAuth metadata discovery. Equivalent to `mcpjam server probe`.
Automated checks
Ran Nov 16, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablemcpjam/inspector is public and reachablegithub.com/mcpjam/inspector ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)No build attestation published for this releaseNot checked
- Open source license (OSI)No licenseFinding
- Commit or release in the last yearLast commit or release 2026-11-14github.com/mcpjam/inspector ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from @mcpjam/cli, @mcpjam/sdkPass
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model17 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubMCPJam is the open-source testing & evaluations platform for MCP server developers who ship. Interactively test tools, prompts, resources, and authorization across 16 client configurations and 170+ models. Run evals on test cases, get insights on what's breaking, and gate regressions to secure every deployment.
🚀 Quick start
Open the hosted app. No install needed.
... or run MCPJam locally for HTTP/S and local STDIO servers:
npx @mcpjam/inspector@latest
👋🏽 Why MCPJam
Clients like ChatGPT, Claude, and Cursor all read your server differently. MCPJam catches those differences, shows where your server breaks, and tells you how to fix it:
- Inspect & debug: Explore your server's tools, resources, and prompts in one place. Test prompts from real LLMs, with full visibility & traces on every JSON-RPC message and OAuth exchange.
- Cross-client evals: Score agent behavior across 16 client configurations — ChatGPT, Claude, Cursor, Copilot, and more — to understand how your MCP integrations perform across all users. Track accuracy, latency, and tool-call performance over time.
- Security & reliability: Automate conformance & behavior checks in CI/CD to catch regressions before they reach production.
Trace view: every tool call, agent step, and JSON-RPC message in one timeline.
🔨 Features
| Capability | What it does |
|---|---|
| Playground | Cross-client chat interface that emulates UIs, tool calls, and skills (with full traces). OpenAI Apps SDK and MCP app UIs, text tools, and a Chrome DevTools-style widget emulator. Read more |
| OAuth Debugger | Visualize your OAuth & EMA requests step-by-step to find the source of errors. Guided MCP OAuth conformance checks across protocol versions 03-26, 06-18, 11-25, and 2026-07-28; DCR, client pre-registration, and CIMD. Read more |
| Server Debugging | Manually run tools, resources, templates, prompts, and elicitation flows with full JSON-RPC logs. |
| Skills | Extend models with reusable behaviors in Playground. Local skills are read from your filesystem and never leave your machine; a project can also carry hosted skills, available on accounts where that is enabled. Read more |
| Workspaces | Shared server groups with real-time team sync, so everyone tests the same config. Read more |
| Evals | Test cases with expected tool calls, run across LLMs, with accuracy metrics over time. Read more |
| CLI | Probe servers, run doctor checks, exercise OAuth, and list tools/resources/prompts from your terminal. Read more |
| SDK | Programmatically drive inspections, snapshot capabilities, and assert on tool/resource shapes from your own tests. Read more |
| CI/CD | Run conformance, E2E tests, evals, and OAuth checks on every PR in GitHub Actions or any pipeline. Read more |
Playground
Debug your server against a model using tool calls or in-panel chat, with Chat, Trace, and Raw views. Supports the OpenAI Apps SDK and MCP app UIs, text tools, and a Chrome DevTools-style widget emulator so you can iterate on widgets locally.
- Invoke a tool to render its widget instantly, or drive your server with an LLM.
- Watch all JSON-RPC and
window.openaimessages in the logs. - Switch the emulator between Desktop, Tablet, and Mobile.
- Test locale changes, CSP permissions, light/dark mode, hover & touch, and safe-area insets.
OAuth Debugger
Guided MCP OAuth conformance checks with step-by-step explanations. Test against every version of the OAuth spec (2025-03-26, 2025-06-18, 2025-11-25, and the 2026-07-28 draft), with support for client pre-registration, Dynamic Client Registration (DCR), and Client ID Metadata Documents (CIMD).
Server Debugging
Everything you need to test an MCP server by hand: run tools, resources, resource templates, prompts, and elicitation flows with full JSON-RPC observability. Every feature of the original inspector, and more.
Skills
Use Skills in Playground to extend models with reusable behaviors. Local skills are read from your filesystem and never leave your machine; a project can also carry hosted skills, available on accounts where that is enabled. Read more
Workspaces
Group servers into shared workspaces with real-time team sync, so everyone on your team tests against the same configuration. Read more
Evals
Define test cases with expected tool calls and run them across multiple LLMs. Track accuracy over time to catch regressions early and improve your server with every iteration. Read more
CLI
Run MCPJam from the terminal for fast local dev loops and CI. Probe servers, run OAuth checks, inspect tools and resources, and execute evals without leaving your shell. Read more
SDK
Programmatic access to MCPJam for custom tooling, scripting, and integrations. Build your own workflows on top of MCPJam's inspection and evaluation primitives. Read more
CI/CD
Wire MCPJam into GitHub Actions, GitLab CI, or any CI system to run conformance, E2E tests, and evals on every PR — and catch MCP server regressions before they ship. Read more
🔌 Install
MCPJam Inspector runs three ways: a hosted web app, a desktop app for Mac and Windows, or via your terminal. The web app is HTTPS-only and has no install. Terminal and Desktop support HTTP/S and local STDIO servers.
Node.js 20+ is only required for the terminal install (
npx). The hosted and desktop apps have no local runtime requirements.
Hosted web app: Open app.mcpjam.com. No install, always latest, and you can share server links with teammates like a Google Doc. HTTPS server URLs only; no STDIO, tunneling, skills, or tasks (those need the local inspector). Hosted docs
Desktop app: Download the installer. Supports HTTP/S and local STDIO, no Node.js required. Install Mac · Install Windows
Terminal: Supports HTTP/S and local STDIO. Open the printed localhost URL after it starts.
npx @mcpjam/inspector@latest
The launcher opens a private access link automatically. To sign in from another browser, use the link printed in your terminal. Docker users can retrieve it with docker logs <container>. Keep it private; a new link is generated on restart unless MCPJAM_SESSION_TOKEN is configured. See the network access instructions for remote addresses and remapped ports.
Docker: Bound to localhost for security. Available at http://127.0.0.1:6274. Always use -p 127.0.0.1:6274:6274 (not -p 6274:6274) to keep the inspector local-only. On macOS/Windows, reach host MCP servers via http://host.docker.internal:PORT.
docker pull mcpjam/mcp-inspector
docker run -p 127.0.0.1:6274:6274 mcpjam/mcp-inspector
👨💻 Contributing
We're grateful you're considering it. Read the contributing guide to get started, and come say hi in Discord.
🌍 Community & links
Website · Blog · Pricing · Docs · Discord · 𝕏 (Twitter) · LinkedIn · YouTube · Instagram · TikTok
📄 License
Licensed under the Apache License 2.0. See LICENSE.
MCPJam MCP questions
How do I add MCPJam MCP to Claude Code?
Run claude mcp add inspector -- npx -y @mcpjam/cli mcp in your terminal. Add --scope user to make it available in every project.
How do I add MCPJam MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What tools does MCPJam MCP provide?
It exposes 17 tools, including connect_server, disconnect_server, list_servers, server_info, ping_server.
How do I run MCPJam locally?
Run npx @mcpjam/inspector@latest in your terminal; it supports HTTP/S and local STDIO servers and prints a localhost URL to open.
Can I test MCP servers across different AI clients?
Yes, MCPJam scores agent behavior across 16 client configurations including ChatGPT, Claude, Cursor, and Copilot.
Does MCPJam help debug OAuth flows?
Yes, its OAuth Debugger visualizes OAuth requests step by step and runs guided conformance checks across multiple protocol versions.
Can MCPJam run evals in CI/CD?
Yes, you can wire MCPJam into GitHub Actions, GitLab CI, or any CI system to run conformance, E2E tests, and evals on every pull request.
Is MCPJam free and open source?
The project is open source and licensed under the Apache License 2.0.