Session Lens - MCP Observability for AI Agents

If an entrepreneur "adopted" this product today, here's the realistic math.

Fermi summary

If you land 80 paying teams at $80/mo that's $77k ARR - 13% chance of hitting it in year 1, and after build costs your expected year-1 take-home is -$15k, so this is a year-2+ bet on a market that may get eaten by incumbents before you get there.

Market size (TAM)

$38.0M

~15,000 companies building production AI agents with tool-use (MCP or equivalent) × $2,500/yr avg observability subscription - nascent but fast-growing segment as of 2026

Year-1 ARR range

$11k - $310k

midpoint $72k

Gross margin

82%

Investment to production

$24k

Dev: $13k for auth, billing, data retention policies, and SDKs for Python/TypeScript. Infra: $4k for production-grade ClickHouse or similar

Probability of success

13%

P(reaching mid case in 12 months)

Expected take-home Y1

$-15300

probability-weighted, after investment

Go-to-market motion

Open-source MCP logging SDK on GitHub as loss-leader → community-driven inbound from AI developer forums/Discord → convert 5-8% to paid dashboard at $79-149/mo per org.

Key risks

Anthropic ships native MCP observability inside Claude.ai or the MCP SDK itself, instantly commoditizing the core value prop
LangSmith, Langfuse, or Honeycomb extend their existing LLM tracing to cover MCP tool calls specifically - they already have the customer base and infra
MCP developers have a strong 'log to stdout and parse it yourself' culture; willingness-to-pay is unproven and conversion from free OSS to paid may be <2%
Data residency concerns: enterprise teams won't route agent telemetry (which may contain PII or proprietary prompts) through a third-party SaaS without SOC2, killing upmarket deals

Generated by the Wishdeal Factory financial-analysis agent. Numbers are honest Fermi estimates, not guarantees. Real outcomes depend on the operator. The studio is bullish on the engineering quality, agnostic on the business outcome.