MCP Server

The Whawit MCP (Model Context Protocol) server lets AI assistants read your observability data, search your team’s operational memory, investigate incidents and queue follow-up work — with the same access your Whawit account has.

This page is about WHAWIT as an MCP server for your AI assistants. For the other direction — connecting your organization’s MCP server so WHAWIT can resolve incidents by acting on your internal systems — see MCP Actions.

Overview

MCP is an open protocol that allows AI assistants to interact with external tools and data sources. The Whawit MCP server exposes thirteen tools covering:

Project discovery and project information
Incidents, analyses, evidence and log excerpts already on record
On-demand analysis of a project’s errors, latency and health
AI incident copilot: root cause, similar incidents, postmortem draft, stakeholder summary — reading the stored results and generating them are separate tools
Operational Memory search (runbooks, docs, learnings, architecture notes)
Fix context and AI-ready fix prompts for the code behind an incident
Ticket creation, synced to Jira

Hosted Server

Whawit provides a hosted MCP server that requires no local installation:

Endpoint	Transport	Use it for
`https://mcp.whawit.ai/mcp`	Streamable HTTP	Recommended. Claude, Claude Code, Cursor and any current MCP client.
`https://mcp.whawit.ai/sse`	HTTP+SSE (legacy)	Clients that only support the older SSE transport.

The hosted server is recommended for most users. It stays up-to-date automatically and requires no maintenance. Customers on a dedicated Whawit instance use that instance’s MCP hostname instead — your Whawit contact has it.

Quick Setup

Claude
Claude Code
Cursor
Other clients

In Claude (web or desktop app), open Settings → Connectors and choose Add custom connector:

Name: Whawit
Remote MCP server URL: https://mcp.whawit.ai/mcp
Add it, then click Connect and sign in to Whawit in the browser window that opens.

claude mcp add --transport http whawit https://mcp.whawit.ai/mcp

Then run /mcp inside Claude Code to complete the sign-in.

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "WHAWIT": {
      "url": "https://mcp.whawit.ai/mcp"
    }
  }
}

Point any MCP client that supports remote servers at https://mcp.whawit.ai/mcp (Streamable HTTP). If your client only speaks the legacy SSE transport, use https://mcp.whawit.ai/sse instead — the server accepts both.

Signing In

There is nothing to copy into your configuration. The URL is the whole setup. When your client first calls the server, Whawit answers 401 with a WWW-Authenticate challenge pointing at its OAuth metadata. Your client reads that, opens your browser, and you sign in to Whawit the way you normally would. The client stores the resulting token and refreshes it on its own.

You are connected when your client lists the Whawit tools — try asking it to “list my Whawit projects”.

You see the same projects your Whawit account sees. Nothing to scope, and no key to rotate or leak.

Available Tools

The MCP server exposes thirteen tools. Each one carries the MCP tool annotations (title, readOnlyHint, destructiveHint, idempotentHint, openWorldHint) that clients use to decide when to ask you before running it. Reading and writing are separate tools: ten tools only read from Whawit and run without a confirmation prompt in Claude; three create something on the Whawit side (an analysis, an AI deliverable, a queued ticket) and Claude asks before running them. None of the tools modifies or deletes existing data.

Annotations at a glance

Tool	Title	`readOnlyHint`	`destructiveHint`	`idempotentHint`	`openWorldHint`
`whawit_list_projects`	List projects	true	—	—	false
`whawit_project_info`	Get project info	true	—	—	false
`whawit_devops_analysis`	Get project incidents and analyses	true	—	—	false
`whawit_run_analysis`	Run on-demand analysis	false	false	false	false
`whawit_get_incident_full`	Get full incident	true	—	—	false
`whawit_get_analysis_full`	Get full analysis	true	—	—	false
`whawit_incident_copilot`	Get incident copilot results	true	—	—	false
`whawit_incident_copilot_generate`	Generate incident copilot results	false	false	false	false
`whawit_knowledge_search`	Search operational memory	true	—	—	false
`whawit_get_fix_context`	Get fix context	true	—	—	false
`whawit_generate_fix_prompt`	Generate fix prompt	true	—	—	false
`whawit_codebase_analysis_prompt`	Generate codebase analysis prompt	true	—	—	false
`whawit_create_ticket`	Queue tracking ticket	false	false	false	true

destructiveHint: false on the three write tools means additive only: they create a new record and never change or remove an existing one. openWorldHint: true on whawit_create_ticket reflects that the ticket leaves Whawit for Jira.

whawit_list_projects

List the Whawit projects your account can access, with their URNs, names and aliases (monitor and integration names). This is how a project you refer to by name (e.g. “acme”) resolves to the URN the other tools take. This tool takes no parameters. Read-only.

whawit_project_info

Get information about a project: its connected integrations, monitors, health status and linked repositories.

string

required

The project URN to get info for.

Read-only.

whawit_devops_analysis

Read-only view of a project’s errors, incidents, failures, latency and health from what Whawit already has on record: the project’s open incidents (with their AI triage) and its most recent analyses, continuously produced from your connected observability providers. Nothing is run or stored. When nothing is on record yet, the response says so and names whawit_run_analysis.

string

required

The question being investigated (e.g., “redis errors”, “API failures”, “high latency”).

string

required

The project URN to look at.

string

default:"concise"

Output verbosity: concise or detailed.

Read-only. Example prompt:

What is failing in project urn:project:abc123 right now?

whawit_run_analysis

Run a new on-demand analysis of a project’s logs for a specific question and an optional time range, using one of the project’s connected log providers. The result (summary, evidence, business impact) is returned when the analysis completes within about 75 seconds; otherwise the analysis URN is returned for whawit_get_analysis_full.

string

required

What to analyze (e.g., “redis connection errors”, “5xx on /checkout”).

string

required

The project URN to analyze.

string

Log-provider integration URN to analyze. Needed only when the project has several log providers — the response lists them when a choice is required.

string

Time range start (e.g., 15m, 1h, 24h). Omit to let Whawit infer the window from the query text.

string

Time range end (ISO date or now); only meaningful together with from.

string

default:"concise"

Output verbosity: concise or detailed.

Creates data: the analysis is stored in the project’s analysis history. Claude asks before running it. Example prompt:

Check for errors in the last 15 minutes in project urn:project:abc123

whawit_get_incident_full

Get the full details of an incident (triage, timeline, Jira/Opsgenie references) together with its correlated analysis when one exists.

string

required

The project URN to fetch incidents from.

string

Specific incident URN to fetch.

string

Filter by severity: critical, high, medium, low. Gets the first matching open incident.

boolean

default:"true"

Include the correlated analysis if available.

Read-only.

whawit_get_analysis_full

Get the complete data of an analysis as formatted markdown and/or the full JSON payload: summary, evidence, business impact and log excerpts.

string

required

The URN of the analysis to fetch (e.g., urn:history:abc123).

string

default:"both"

Output format: markdown, json, or both.

boolean

default:"false"

Include raw log events in output.

number

default:"20000"

Maximum output length in characters.

Read-only.

whawit_incident_copilot

Read-only: returns the AI deliverables Whawit has already stored for an incident — probable root cause with evidence, similar past incidents and how they were resolved, a non-technical status update, or the postmortem. Nothing is generated; when no result of that kind exists yet, the response says so and names whawit_incident_copilot_generate.

string

required

The incident URN.

string

required

root-cause, similar, postmortem or stakeholder-summary.

Read-only.

whawit_incident_copilot_generate

Generate an AI deliverable for an incident on the Whawit side and store it on the incident: probable root cause with evidence (root-cause), similar past incidents (similar), a non-technical status update for a given audience (stakeholder-summary), or a postmortem draft (postmortem). Generation takes 1–2 minutes. Without regenerate, the stored deliverable is returned when one already exists.

string

required

The incident URN.

string

required

root-cause, similar, postmortem or stakeholder-summary.

string

default:"exec"

Only for stakeholder-summary: exec, customer or technical.

boolean

default:"false"

Generate a new deliverable even when one is already stored (root-cause, similar, stakeholder-summary; slower). The previous ones are kept.

Creates data: stores AI artifacts on the incident. The postmortem draft is created only when the incident has none — an existing postmortem is never overwritten from here. Claude asks before running it.

whawit_knowledge_search

Search the team’s Operational Memory (runbooks, docs, incident learnings, procedures, architecture notes) for a project. Every hit carries a citation URN.

string

required

The search query.

string

required

The project URN to search in (e.g., urn:project:abc123).

array

Optional filter: runbook, doc, convention, architecture, investigation-guide, learning, procedure.

boolean

default:"false"

Include draft (unapproved) knowledge items.

Read-only. Example prompt:

Search for how authentication works in project urn:project:abc123

whawit_get_fix_context

Everything needed to fix an incident’s underlying code locally: the correlated analysis with evidence and log excerpts, the engineering actions, prior automated fix runs, and the project’s GitHub repositories (clone URL and default branch).

string

Incident URN to build the fix context from (resolves its correlated analysis).

string

Analysis (history) URN to build the fix context from directly.

One of the two is required. Read-only.

whawit_generate_fix_prompt

Generate an AI-ready prompt for fixing the issues identified in an analysis.

string

required

The URN of the analysis to generate a fix prompt for.

string

default:"all"

Focus area: code, infrastructure, or all.

Read-only.

whawit_codebase_analysis_prompt

Generate a prompt scaffold for analyzing a codebase: project overview, recent analyses and generic exploration instructions. It does not look up architecture or conventions — use whawit_knowledge_search for those.

string

required

The project URN to analyze.

Read-only.

whawit_create_ticket

Queue the creation of a tracking ticket (synced to Jira when a Jira integration is connected) from an analysis or incident.

string

Analysis (history) URN to create the ticket from.

string

Incident URN — its correlated analysis will be used.

string

Optional instructions to steer the ticket content.

One of historyUrn or incidentUrn is required. Creates data: the call is asynchronous — the response confirms the ticket was enqueued and does not include a ticket ID.

MCP Resources

The server also exposes projects as MCP resources:

whawit://project/{urn}

Reading a project resource returns the project document (name, description, URN and settings) as JSON.

Authentication

The server authenticates you with OAuth 2.1, discovered automatically. You add a URL; your client does the rest. An unauthenticated request is answered with an RFC 9728 challenge that points at the metadata for the endpoint you connected to:

HTTP/2 401
WWW-Authenticate: Bearer resource_metadata="https://mcp.whawit.ai/.well-known/oauth-protected-resource/mcp"

The client follows that to the metadata documents, registers itself, sends you to your browser to sign in, and then calls the server with a Bearer token. Every MCP client that implements the authorization spec — Claude, Cursor, VS Code, Windsurf — does this without being told.

Headless environments

A CI job or a script has no browser to open. For those, the server also accepts an X-API-Key header, which you can generate in the Whawit web app under Settings > User Settings. Prefer OAuth anywhere a person is present: a key is something to store, rotate and eventually leak.

Transport Protocols

Protocol	Endpoint	Notes
Streamable HTTP	`/mcp`	Recommended. Stateless: every request is self-contained, no session header is required, and requests may be served by any instance of the service.
HTTP+SSE (legacy)	`/sse`	Kept for clients that predate Streamable HTTP. The connection is bound to the instance that accepted it.

Both endpoints support the full set of tools and resources.

Example Workflows

Whawit MCP tools in action — Using Whawit MCP tools in Cursor to analyze incidents

Investigating an Incident

Ask your assistant to list your projects:
“List my Whawit projects”
Get incident details:
“Get the latest critical incident from the acme project”
Ask the copilot for a root cause (the stored result is returned when one exists; otherwise Claude offers to generate it):
“What is the probable root cause of that incident, with evidence?”
Pull the fix context and let the assistant work on the code:
“Get the fix context for that incident and propose a fix”
Track it:
“Create a ticket for it”

Searching Operational Memory

“Search the project knowledge base for how the payment processing flow works in urn:project:payments”

Running an On-Demand Analysis

“Run an analysis of Redis connection errors in the last 30 minutes for urn:project:backend”

Troubleshooting

Tools not appearing in your client

Check the URL ends in /mcp (or /sse for a legacy SSE client)
Verify your MCP configuration file syntax is valid JSON, if your client uses one
Restart the client after configuration changes
Ensure mcp.whawit.ai is reachable from your network

Authentication errors

Check that your configuration has no headers block — an old X-API-Key entry left in place takes precedence over OAuth and will fail once that key is revoked.
Sign out and reconnect so the client runs the browser flow again; tokens expire.
Confirm the account you signed in with has access to the project you are asking about — try “list my Whawit projects” to see what it can reach.

Project not found errors

Verify the project URN format is correct (urn:project:...)
Ensure you have access to the project in Whawit
Run whawit_list_projects to see available projects

Next Steps

VS Code Extension

Install the Whawit extension for integrated observability in your editor.

GitHub Copilot

Use @whawit in GitHub Copilot Chat for AI-assisted analysis.

Getting started

On-call

MCP Actions

Policies

Concepts

Delivery reliability

Developer Tools

Demos & resources

Company

Overview

Hosted Server

Quick Setup

Signing In

Available Tools

Annotations at a glance

whawit_list_projects

whawit_project_info

whawit_devops_analysis

whawit_run_analysis

whawit_get_incident_full

whawit_get_analysis_full

whawit_incident_copilot

whawit_incident_copilot_generate

whawit_knowledge_search

whawit_get_fix_context

whawit_generate_fix_prompt

whawit_codebase_analysis_prompt

whawit_create_ticket

MCP Resources

Authentication

Transport Protocols

Example Workflows

Investigating an Incident

Searching Operational Memory

Running an On-Demand Analysis

Troubleshooting

Next Steps

VS Code Extension

GitHub Copilot

​Overview

​Hosted Server

​Quick Setup

​Signing In

​Available Tools

​Annotations at a glance

​whawit_list_projects

​whawit_project_info

​whawit_devops_analysis

​whawit_run_analysis

​whawit_get_incident_full

​whawit_get_analysis_full

​whawit_incident_copilot

​whawit_incident_copilot_generate

​whawit_knowledge_search

​whawit_get_fix_context

​whawit_generate_fix_prompt

​whawit_codebase_analysis_prompt

​whawit_create_ticket

​MCP Resources

​Authentication

​Transport Protocols

​Example Workflows

​Investigating an Incident

​Searching Operational Memory

​Running an On-Demand Analysis

​Troubleshooting

​Next Steps

VS Code Extension

GitHub Copilot

Overview

Hosted Server

Quick Setup

Signing In

Available Tools

Annotations at a glance

whawit_list_projects

whawit_project_info

whawit_devops_analysis

whawit_run_analysis

whawit_get_incident_full

whawit_get_analysis_full

whawit_incident_copilot

whawit_incident_copilot_generate

whawit_knowledge_search

whawit_get_fix_context

whawit_generate_fix_prompt

whawit_codebase_analysis_prompt

whawit_create_ticket

MCP Resources

Authentication

Transport Protocols

Example Workflows

Investigating an Incident

Searching Operational Memory

Running an On-Demand Analysis

Troubleshooting

Next Steps