Docs › Reference
MCP tools
Every tool and ready-made prompt Relia's MCP server gives your AI assistant, with arguments, confirmations and plan notes.
The tools your AI assistant gets once it is connected to Relia's MCP server at https://devreliaapi.tryrelia.com/mcp (see Connect your AI assistant). Every tool acts as you: it only reads projects you can see, with your role in each.
| Rule |
Details |
project argument |
Every tool except relia_projects and relia_create_project takes project: the project's id or name (up to 100 characters). It is optional when the connection reaches only one project. In a repository set up with Relia, the id is RELIA_PROJECT_ID in .env.local, or project_id in src/relia.ts / relia.tracing.js. |
| Read-only by default |
Every tool only reads, except the five marked "Yes" under Confirm. |
confirm: true |
Actions do nothing until called with confirm: true. Without it, Relia answers with the question to ask you first. Your assistant should only set it after you say yes. |
| Audit log |
Every call is recorded in the project's audit log: reads as "AI assistant read Relia (MCP)", actions under their own name. |
| Limit |
300 calls an hour per token or sign-in. |
| Privacy |
Answers never include who a visitor was, what they typed, or customers' names and e-mails (only counts and plan names). |
Your projects and what needs you
| Tool |
What it returns |
Arguments |
Confirm |
Plan notes |
relia_projects |
The projects this connection can read: id, name, plan, your role, number of services. |
None |
No |
All plans |
relia_status |
Relia's verdict: what needs you, each area's state (uptime, APIs, database, servers, page speed, users, AI agents) and the incidents that need a fix. |
project |
No |
All plans |
relia_list_incidents |
Incidents, most important first: id, severity, users affected, fix state. |
project, status (open default, or resolved), limit (1 to 25, default 10) |
No |
All plans |
relia_get_incident |
One incident: what happened, impact, what to do next, how it unfolded, evidence per layer with example trace ids, deploys around then, and the signed-in customers it hit (counts and plans only). |
project, incident_id (required) |
No |
Free: the headline and size only, with a note that the explanation is on Growth and Pro. |
relia_get_fix_plan |
Steps to fix an incident: root cause, files and lines, the change, how to verify. |
project, incident_id (required) |
No |
Growth and Pro. Uses one of the month's fix plans the first time (Growth 30, Pro 50), then the stored plan is reused. Free answers with a plan message. |
relia_ask |
An answer about your product's live data, citing its evidence. |
project, question (required, up to 1,000 characters) |
No |
Uses the month's Ask Relia allowance (Free 30, Growth 100, Pro 300). |
Evidence
| Tool |
What it returns |
Arguments |
Confirm |
Plan notes |
relia_trace |
One request end to end: every span in order (services, database calls, outgoing calls) with durations, what failed and the error message. |
project, trace_id (required; from incidents, logs or database results) |
No |
Traces are kept for your plan's history (7, 30 or 90 days). |
relia_logs |
Warning and error log lines grouped by message pattern (count, last seen, code location, example traces). With pattern, that pattern's latest lines. |
project, search, service, level (warn or error), pattern, range (15m, 1h, 6h, 24h default, 7d, 30d) |
No |
All plans |
relia_session |
One visit step by step: pages, clicks, failed requests and errors with their messages, rage clicks and stuck moments. Never who the visitor was or what they typed. |
project, session_id (required; from an incident's evidence) |
No |
All plans |
relia_database |
Database queries grouped without values, most total time first: calls, p95, failures, example slow or failed traces; plus N+1 patterns. |
project, service, range (15m, 1h, 6h, 24h default, 7d, 30d) |
No |
All plans |
relia_vitals |
Page speed for real visitors: LCP, INP, CLS (and FCP, TTFB) at p75 with ratings, the change vs the period before, and the worst pages. |
project, range (24h, 7d default, 30d) |
No |
All plans |
relia_uptime |
Uptime monitors (up or down now, uptime %) and Relia's own checks of key pages and APIs (passing, failing step). |
project, range (24h default, 7d) |
No |
All plans |
relia_changes |
Recent deploys Relia saw, changes reported to Relia, and incidents that began, newest first. |
project, hours (1 to 336, default 48) |
No |
All plans |
Dashboards and funnels
| Tool |
What it returns |
Arguments |
Confirm |
Plan notes |
relia_dashboards |
The team's dashboards with their ids, and Relia's dashboard per service with why each is worth opening. |
project |
No |
All plans |
relia_dashboard |
One team dashboard read out panel by panel: the number now and its change, top items, monitors, checks and promises, open incidents, and what moved after a deploy or incident. |
project, dashboard_id (required), range (15m, 1h, 6h, 24h, 7d, 30d; default the dashboard's own), service |
No |
All plans |
relia_funnels |
The journeys Relia follows (checkout, sign up…): how many people finished in the last 7 days vs the week before, the step where most are lost, and goals no funnel follows yet. |
project |
No |
Why people leave each step: Growth and Pro. |
Setup
| Tool |
What it returns |
Arguments |
Confirm |
Plan notes |
relia_setup |
Exact steps to install Relia in the code open in your editor, for this project, including two choices for you (screen replays, identifying users). |
project |
No |
Owners, admins and members (not viewers): the steps include the project key. |
relia_check_setup |
What has arrived in Relia: requests per service, browser sessions, source maps, identified users, AI agent runs, each with the step left. |
project |
No |
All plans |
Actions (asks you first)
| Tool |
What it does |
Arguments |
Confirm |
Plan notes |
relia_mark_fixed |
Tells Relia an incident is fixed. Relia watches real traffic and confirms the fix worked, or says it is still happening (usually 15 to 30 minutes after the fix is live). |
project, incident_id (required), note (what changed, up to 500 characters), confirm |
Yes |
Any member. Only for an open incident. |
relia_set_verdict |
Teaches Relia about an incident: correct (real), wrong (false alarm), expected (by design, stop raising it), mute, or bug (always needs a fix). |
project, incident_id (required), verdict (required), note (up to 1,000 characters), confirm |
Yes |
expected, mute and bug need an owner or admin. |
relia_ack |
"I'm on it": stops paging the next person on call for this alert and acknowledges PagerDuty. |
project, incident_id (required), confirm |
Yes |
Not viewers |
relia_report_change |
Records a deploy, config change or feature flag Relia can't see, so it links problems to it. A deploy starts fix verification. |
project, service, kind (deploy default, config, flag, other), version, previous_version, note (up to 500 characters), confirm |
Yes |
Not viewers. Same as the Report changes API. |
relia_create_project |
Creates a Relia project for a repository that has none, then points to relia_setup. |
name (required, up to 100 characters), timezone (optional, for the daily summary), confirm |
Yes |
Needs a connection that covers all your projects. At most 5 an hour. Your account's plan rules apply. |
Prompts
Ready-made requests your tool lists (in Claude Code, for example, /mcp__relia__fix-top).
| Prompt |
Title |
What it asks your assistant to do |
setup |
Set up Relia |
Call relia_setup, ask you its two questions first, show every change before writing it, never commit, then check with relia_check_setup after you start the app. |
check |
Is Relia receiving everything? |
Call relia_check_setup and say what Relia receives and the one thing to do next. |
status |
What needs my attention? |
Call relia_status, and get the fix plan for anything that needs a fix. |
fix-top |
Fix the most important problem |
Pick the most important incident, read its evidence and fix plan, find the cause in your repository, show you the change before writing it, never commit, then offer relia_mark_fixed after you deploy. |
deploy-check |
Did my last deploy break anything? |
Find the latest deploy with relia_changes, then compare incidents, error logs and database queries before and after it. |
review-pr |
Review this change against production |
Check the routes, queries and pages in your current git diff against production with relia_ask, relia_database, relia_vitals and relia_list_incidents, and list the risks. |
Next
Last checked against the product: 1 Nov 2026