MCP installation

Install Baseline, run a check, accept a clean run.

Baseline is a local CLI and MCP server for coding-agent workstation health. It runs probes against your approved agent target, writes local reports, and compares future runs against the clean baseline you explicitly accept.

Setup -> run -> accept -> compare
curl -fsSL https://trackbaseline.com/install.sh | sh
baseline --version
baseline doctor
baseline setup
baseline run --mode fast
baseline report
baseline accept RUN_ID --confirm "accept RUN_ID" --label clean-local
baseline compare

Universal MCP smoke

Run this before debugging a client-specific plugin. It lists the local MCP tools without starting an agent eval.

Universal MCP smoke
printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | baseline serve mcp

Client setup paths

ClientRegisterVerify
Codex pluginInstall the Baseline plugin from the local marketplace after the CLI is on PATH.baseline --version, then plugin tools include Baseline.
Hermes native MCPAdd baseline with command baseline serve mcp under mcp_servers.hermes mcp list and hermes mcp test baseline.
OpenClawInstall the plugin or configure manual stdio MCP with baseline serve mcp.openclaw mcp list after restarting the gateway.

Side effects: baseline doctor is read-only. baseline setup, baseline run, and scheduled runs send the configured probe messages and write local Baseline artifacts.

Distribution

The installer downloads the latest checksummed release asset for macOS or Linux from GitHub Releases, verifies checksums.txt, and installs baseline into ~/.local/bin by default. Set BASELINE_INSTALL_DIR for a different destination or BASELINE_VERSION for a pinned release.

Pinned or custom install
curl -fsSL https://trackbaseline.com/install.sh | BASELINE_INSTALL_DIR=/usr/local/bin sh
curl -fsSL https://trackbaseline.com/install.sh | BASELINE_VERSION=v0.1.0 sh

Operator guides

Guide / 2026-06-01How to run a coding agent health check before work drifts.

A practical coding agent health check for memory, repo awareness, MCP visibility, latency, safety, and style drift.

Read →
Guide / 2026-06-01Run an MCP server health check without adding tool sprawl.

Verify MCP server visibility, tool count, setup, scrub preview, and recovery paths for local coding agents.

Read →
Resource / 2026-06-01The MCP debugging cheatsheet for agent workstations.

A cheatsheet for debugging missing MCP tools, broken local CLI setup, and coding-agent server drift.

Read →

Cloud sync

Cloud sync is optional. The local product works without an account; Pro adds redacted history, workspaces, retention, and remote MCP account operations after a local runner is already producing scrubbed summaries.

Optional Pro sync
baseline sync on --url https://trackbaseline.com --token YOUR_BASELINE_TOKEN
baseline doctor
baseline sync push

Remote MCP

Pro accounts can also connect to the cloud MCP at https://trackbaseline.com/mcp. The remote MCP never runs local probes; it reads account history, hotspots, self-history comparisons, workspace tokens, and Stripe portal handoffs after magic-link session auth.

Remote MCP sequence
POST https://trackbaseline.com/api/auth/magic-link
POST https://trackbaseline.com/api/auth/consume
Authorization: Bearer YOUR_SESSION_TOKEN
POST https://trackbaseline.com/mcp

Safety model

The MCP can read what the connected agent gives it. Baseline defaults to local SQLite and redacted summaries. Raw outputs are not exported unless allow_raw_output is enabled in ~/.baseline/config.json.

Recommended first Good Baseline

Run the sequence once before important work. If the report shows the wrong workspace, missing MCP tools, unsafe scrub output, or surprising latency, repair the workstation before accepting the run.

First Good Baseline
baseline setup
baseline run --mode fast
baseline report RUN_ID
baseline accept RUN_ID --confirm "accept RUN_ID" --label clean-local
baseline compare