Namespace guard for merged toolboxes — by TextForAgents LLC. Before your fleet connects a new MCP server, audit its tool names against your toolbox — exact and normalised, trust-graded, colliding publisher named — then merge publisher-bound, with a signed routing receipt for every call on a name that ever collided.
The MCP specification has no formal namespace mechanism: if two servers registered to the same agent expose the same tool name, disambiguation becomes impossible. Microsoft Research counted name collisions between 775 tools across a 1,470-server catalog — search alone appears across 32 distinct servers — and hosts today either prefix inside their own client or refuse the merge outright; neither audits a toolbox before connection, grades a shadowing claim, or records which server actually received a call. This service is that guard. Its pre-connection collision audit grades a candidate server's tool names against your fleet's existing toolbox in exact AND normalised forms (case, separator, prefix-stripped, cross-level) as CLEAR / COLLIDES / SHADOWS-TRUSTED (an untrusted publisher claiming a name a trusted server owns) / INVALID, with the colliding publisher named. Its deterministic merge binds every tool as <publisher>__<tool> and refuses a bound-name duplicate rather than overwriting it. Its Router emits an Ed25519-signed, hash-chained routing receipt for every call on a name that ever collided — which server, under which resolution rule, with an args-hash — so "the agent called search" becomes an attributable fact. Its claimed-names census keeps which publishers claim which common verbs current across the server population.
Prices: Verdict query $0.01/query · Pre-connection collision audit $4.99/server · Routing receipts add-on $9.99/fleet/mo (5,000 receipts included, then $0.01 each) · Fleet pass $19.99/mo · Organisation pass $49.99/mo (x402). One free collision audit, one time per operator.
Stated plainly: the fixture census rate (0.967) is a fixture artifact, not a market figure — the live top-N overlap rate is unmeasured, and the census is the instrument that measures it; a routing receipt settles attribution, not content disputes (raw args are never in a receipt); receipt verification detects tampering only against a pinned chain head; willingness to pay is unmeasured.
API: https://textforagents.com/products/namespace-guard-for-merged-toolboxes/ · Support: support@textforagents.com