MCP server

The Nijam MCP server lets any Model Context Protocol-capable agent - Claude Code, Cursor, Codex, Claude Desktop, and others - answer questions about your test runs straight from your real history:

“Why is the checkout suite red?” · “Is cart updates quantity flaky?” · “When did this test start failing?”

The agent calls the server, the server reads your data from the Nijam API, and the agent reasons over the result. It is read-only: it can pull runs, failures, and flakiness, but can never upload, change settings, or touch anything else.

Prerequisites

You need a Nijam read key (nij_rk_…). Read keys are a distinct, read-only credential - separate from the nij_sk_… ingestion keys your CI uses, so a leaked CI key can never expose data and a read key can never pollute it.

Open Secret keys in your organization settings and switch to the Read keys (MCP) tab.
Choose Create key, then copy the nij_rk_… token - it’s shown once.

Read keys are always org-wide, so the agent can resolve any project in the organization by name or slug on its own.

Set up your client

Pick your agent below. In every case, replace nij_rk_... with the read key you created.

Claude Code

claude mcp add -s user nijam -e NIJAM_API_KEY=nij_rk_... -- npx -y @nijam/mcp-server

-s user registers Nijam across all your projects - the default scope only adds it to the current directory, so it won’t show up elsewhere. After adding it, start a new Claude Code session so the tools load, then confirm:

claude mcp list   # → nijam: ✓ Connected

Cursor

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "nijam": {
      "command": "npx",
      "args": ["-y", "@nijam/mcp-server"],
      "env": { "NIJAM_API_KEY": "nij_rk_..." }
    }
  }
}

Codex

Add to ~/.codex/config.toml:

[mcp_servers.nijam]
command = "npx"
args = ["-y", "@nijam/mcp-server"]
env = { NIJAM_API_KEY = "nij_rk_..." }

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "nijam": {
      "command": "npx",
      "args": ["-y", "@nijam/mcp-server"],
      "env": { "NIJAM_API_KEY": "nij_rk_..." }
    }
  }
}

npx -y @nijam/mcp-server fetches and runs the server on demand - there’s nothing to install or keep updated. Point it at a self-hosted API with the optional NIJAM_API_URL env var (defaults to https://api.nijam.dev).

What the agent can do

The server exposes nine read tools. get_projects is the entry point - it lists the projects the key can see; every other tool accepts a project id, slug, or name, so you never paste a UUID.

Tool	Answers
`get_projects`	What projects can I see? (resolves “my project” to an id/slug)
`get_latest_run`	What’s the state of the latest run, optionally on a branch?
`list_runs`	Run history, filterable by status and branch
`get_run`	One run’s stats and per-spec-file breakdown
`list_failing_tests`	Which tests failed in a run, and with what error
`get_failure`	Full detail of one failure: every attempt, error, and artifact links
`is_test_flaky`	A verdict, with evidence, on whether a test is flaky
`get_test_history`	One test across the last 30 runs - spot when it broke
`list_flaky_tests`	The flakiest tests in a project, ranked

Where a run has artifacts (Playwright traces), they come back as short-lived signed links the agent can open, minted on demand.

Security

Read-only. Read keys cannot write - the upload routes reject them - so the MCP server can never create, change, or delete anything.
Scoped. A key sees only its organization’s projects, and anything outside its scope returns a plain 404.
Local. The server is a local subprocess; the key never leaves your machine, and it is never exposed to the model.
Revocable. Revoke a read key anytime from Secret keys - agents using it lose access immediately.