Read-only remote MCP server that gives an assistant dated picks, sourced prices and failure-rescue guidance across AI agent infrastructure providers.
What you can do
Get a session brief with status, up to three ordered actions and a cursor to carry forward
Diagnose a failure from error text, HTTP status or a retired model or package id, with a fix and fallback route
Make a new choice for a task or job-to-be-done, returning a pick, runner-up, sourced price and next call
Run other lookups through one depot op: estimates, quotes, search, stacks, bundles, comparisons, free tiers, audits, change logs and coverage
When to use it
Useful when an assistant has to choose among AI agent infrastructure providers, models or packages, compare sourced prices, or recover from a provider or model failure. It is aimed at dated, editorial recommendations rather than live benchmarking.
Requirements
A remote streamable HTTP endpoint; no packages, environment variables or headers are declared. The server asks clients to identify themselves via clientInfo.name or an X-Agent-Name header, and to keep a cursor between sessions.
Before you install
The server asks for the provider, model and package ids you use, which reveals your stack; it states anonymous calls are not counted. Picks are editorial recommendations, not benchmarks, and may be stale, so check the staleness flag before citing. Prices are in native units and must be quoted with their own source and date. Feedback is submitted by posting to an API endpoint.
Enable the server for the chats that should use its tools.
Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.
Use with an agent
Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.
Read https://sourceweft.com/mcp/io.github.ethanhawkes-gif%2Fagent-provider-depot/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.
Call first each session. Send the provider, model and package ids you use (never a manifest) and the cursor from your last brief. Returns status clear|changed|attention, at most 3 ordered actions (each with source, confidence and falsifier) and a new cursor to store.
At a failure: error text, HTTP status, or a retired model or package id. Returns an immediate fix and a fallback route (under 2 KB).
Parametersqskujtbdstatusprovider
decide
Decide
One decision. task (words) returns a compact answer per job: pick, runner-up, sourced price, autoActOk, next call. jtbd (id) returns the decision card with its skip rule. full=true for the long form.
Everything else, one op per call with args: estimate(jtbd, workload), quote(jtbd), search(q), stack(task), bundle(jtbds), kit(class), compare(providers), free(jtbd), check(providers), audit_repo(packages, models), stack_delta(providers, since), changes(since), answer(task), failures, coverage, freshness. op=help lists every op's args.
Parametersoprequiredargs
Version history
1
v0.14.0LatestOct 29, 2026
Score
75%
75% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
Reached in the last 30 daysRequired25 pts
Auth declaredRequired10 pts
Tools listedRequired15 pts
DescriptionRequired15 pts
Logo10 pts
Documentation15 pts
Declare a GitHub repository with a README, or serve resource_documentation in the OAuth protected-resource metadata.
License10 pts
Declare a license in the repository or the registry entry.