Faros AI MCP server: query delivery metrics and AI spend
Faros joins Git, CI, incident and AI-tool data into engineering metrics. Its hosted MCP server lets an agent query those joined metrics instead of the raw events.
Checked 3 October 2026 · from Faros's 30 April 2026 release post and a request to the hosted endpoint
One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.
Summary
Faros AI runs an official MCP server. It was announced on 30 April 2026, alongside Experiments and expanded AI-tool coverage, and it is hosted at https://prod.api.faros.ai/mcp.
The server answers from data Faros has already normalized and attributed across the delivery pipeline. Faros names lead time, deployment frequency, AI tool adoption, incident rates and token consumption as examples, so an agent gets metrics rather than raw Git or Jira events.
Faros's technical documentation is password-protected, so the tool names and parameters are visible only to customers. Plan the connection on that basis.
How Faros AI works with MCP
Faros is an engineering-intelligence platform. It ingests from source control, CI, incident and AI coding tools, then attributes the results to teams, services and time periods.
The MCP server exposes that joined model. Questions such as token consumption by team, or how PR cycle time moved after a model switch, are answered from Faros rather than reconstructed by the agent from several systems.
Faros describes the server as working with ChatGPT, Claude and Perplexity. It uses OAuth: the endpoint advertises its authorization server through the standard /.well-known/oauth-protected-resource/mcp document.
Every vendor here gets the same four checks: official, maintained, distributed and usable. Faros AI passes 2 of them.
| Check | Result | Evidence |
|---|---|---|
| Official | Pass | Faros announced its own MCP server in its 30 April 2026 release post. |
| Maintained | Unknown | Faros publishes no changelog or version history for the server. |
| Distributed | Pass | Hosted at https://prod.api.faros.ai/mcp. On 3 October 2026 the endpoint answered an unauthenticated request with HTTP 401 and OAuth protected-resource metadata. |
| Usable | Unknown | The tool list is only visible to a signed-in Faros customer, so the callable tools could not be confirmed. |
| Part of the work | What Faros AI has here | Can an agent reach it |
|---|---|---|
| CI & release | Lead time, deployment frequency, PR cycle time | Yes. The joined delivery metrics Faros computes |
| Production observability | Incident rates attributed to teams and services | Partial. Incident metrics, not live telemetry |
| AI cost management | AI tool adoption and token consumption | Yes. Spend and usage by team, as Faros ingests it |
Faros is strongest where a question spans several systems: delivery speed against AI spend, or incidents against deployment volume. For a single pull request or a single pipeline run, the source system's own server is more direct.
What can an AI agent do with Faros AI?
A metric comes back empty for a team you know is active
Faros can only answer from the sources a customer has connected. An AI tool that is not instrumented, or a repository outside the configured scope, produces no data rather than an error.
Guard: Check the connected sources in Faros before treating an empty answer as zero usage.
How to configure Faros AI MCP for agents
- 1
Confirm access first
The server reads your organization's Faros data, so the connecting user needs a Faros account with the relevant role.
- 2
Expect OAuth
The endpoint returns 401 with protected-resource metadata until the client completes sign-in.
- 3
Scope by role
Faros supports custom roles with their own permission sets, which determine what the agent can see.
Paste this into your agent’s instructions
When you answer from Faros data, state: - The metric, the team or service, and the time window. - Which connected sources the metric is computed from. If a metric is empty, say that the source may not be connected. Do not report an empty result as zero usage.
Do you need the MCP server at all?
For raw events, connect the source system instead. GitHub's server returns pull requests and reviews; PagerDuty's returns incidents. Those answer what happened, not how a team's metrics moved.
Faros's value over those servers is the attribution: the same PR, deployment and incident joined to a team and a period. If the question needs that join, the Faros server is the shorter route.
Use Faros for questions across systems and over time. Use the source system for a single record.
Faros AI integration setup
Faros hosts the server, so there is nothing to install. Add the URL to your client and complete the OAuth sign-in.
Claude Code
Add the hosted endpoint over HTTP. The first tool call opens the Faros sign-in.
claude mcp add --transport http faros https://prod.api.faros.ai/mcp
Cursor
Add it to .cursor/mcp.json for one project or ~/.cursor/mcp.json for all of them.
{"mcpServers":{"faros":{
"url":"https://prod.api.faros.ai/mcp"}}}What the vendor documents
Public discussion of the server is limited to Faros's own release post. It gives four scenarios: forecasting AI spend after a vendor price change, combining Faros data with an outside spreadsheet, interrogating experiment results, and building monitoring workflows.
Should you connect an agent to Faros AI?
Yes, if your organization already uses Faros. Best for engineering leaders asking cross-system questions about delivery and AI spend.
- Strongest fit
- Yes, if your organization already uses Faros.
- Main advantage
- Answers come from metrics already joined across Git, CI, incidents and AI tools.
- Main weakness
- The tool list and parameters are behind the customer login.
- Operational risk
- An unconnected source returns empty results that read like zero activity.
From Oodle
One platform for agent traces and infrastructure
Agent traces usually sit in a different product from the rest of your telemetry, so when a slow database makes an agent slow the symptom is in one tool and the cause is in another. Oodle keeps both in one query surface.
See agent observabilityFrequently asked questions
Does Faros AI have an MCP server?
Yes. Faros announced it on 30 April 2026 and hosts it at https://prod.api.faros.ai/mcp. It authenticates with OAuth.
What can an agent ask the Faros MCP server?
Questions about the metrics Faros computes, such as lead time, deployment frequency, AI tool adoption, incident rates and token consumption by team.
Where is the Faros MCP tool list?
In Faros's technical documentation, which is password-protected and available to customers.
Is Faros AI the same as faros.sh?
No. faros.sh is an unrelated infrastructure platform with a similar name.
