Summary: On November 12, 2026, Firecrawl shipped MCP v3, a single MCP connection that exposes the full Alexandria catalog — company data, people data, research, and government records, plus Search, Scrape, and Parse — as 3,000+ tools. The headline mechanism is a waterfall: when the first provider returns “not found,” v3 keeps trying the next one and your agent makes one call and gets one answer. It runs in Claude, ChatGPT, and any harness that speaks MCP.
One connection instead of one integration per provider
In v2-style setups, wiring an agent to data providers meant connecting each one separately. v3 collapses that: the agent describes what it needs, and the MCP selects the right tool from the catalog. Firecrawl states that provider discovery is about 8x faster than in v2, so the agent spends its time on the task rather than the lookup.
This continues the direction we covered earlier in One Endpoint Now Reaches Providers, Not Just Pages — moving provider choice behind a single interface so the caller stops caring which source answers.
The waterfall fixes the “not found” dead end
The concrete problem the announcement opens with: ask an agent for a contact’s work email and it often comes back empty, even though the answer sat with the next provider the agent never asked. The people enrichment waterfall in v3 checks one source, then the next, until it finds the record. That changes the failure mode from a dead end that interrupts the run into a longer single call — the cost moves from your orchestration code into the MCP itself.
More headroom, and subtasks that run while the agent works
Firecrawl reports rate limits up to 60x higher and 4x more usage through the MCP, with search about 2x faster. The stated reason headroom matters: a research agent checking 200 companies in an afternoon “shouldn’t stall at company 40.”
The other structural change is delegation. An agent can spawn background subtasks and keep working while they finish — for example, splitting a request for background on ten prospects into ten parallel subtasks and collecting results as they land.
Every plan includes free MCP usage that refreshes daily, and the endpoint is live today at alexandria.firecrawl.dev/go/mcp.
Our take: For a small team building company or people lookups into an agent, we would connect this now — not because of the 60x or 8x numbers, which are vendor-stated with no methodology attached, but because the setup cost is one MCP connection and the free daily usage makes a test cheap and reversible. Swap in lookups your current stack has failed on and compare find rates against what you get today. Who can safely ignore it: builders whose products never touch external enrichment, or teams already committed to a provider whose find rate they have measured and trust. The position reverses if a real test shows the waterfall rarely finds records the first provider missed — then v3 is just a wrapper around the same answer — or if the free tier’s daily usage binds before your workload fits under it. Also weigh lock-in: once waterfall behavior shapes how your agent handles misses, moving providers means re-tuning that behavior.
The announcement leaves open what we would check before depending on it in production: which providers sit in the people enrichment waterfall, what data terms they carry, and the actual per-plan rate limits behind “up to 60x higher.”
Sources
View the original post on X
