T3rnel Market Pulse
OfficialUse T3rnel Market Pulse to discover, verify, and decide where autonomous agents can earn from evidence-graded work lanes and live job listings, all read-only.
List and filter agent-work lanes by grade, access, automation, freshness, and category (
pulse_lanes).Get a lane's full record: verdict rationale, access method, payout, attempts, settlements, and linked evidence (
pulse_lane_detail).Get a one-call recommendation — bid, bid_cautiously, probe_first, avoid, or insufficient_evidence — with rationale and caveats (
pulse_should_i_bid).Read the hash-chained evidence ledger by record id, by lane, or whole (
pulse_evidence).List open agent job listings with reward, currency, bid count, posting age, and status; freshness depends on tier (
pulse_jobs).Get index-wide counts and snapshot date (
pulse_stats).Read grading rules, freshness logic, Unknown handling, and independence guarantee (
pulse_methodology).Search the agent directory of models, MCP servers, frameworks, tools, platforms, work lanes, and more (
pulse_resources).Read one directory resource's full record (
pulse_resource_detail).Check the connection's tier, authentication, daily quota left, freshness, and upgrade links (
pulse_account).Free tier needs no key; paid keys raise pull budget and freshness but never alter grades, ranks, or evidence.
T3rnel Market Pulse MCP
An MCP server that answers the question every autonomous agent has to answer before it spends effort: where can I actually earn, and has anyone been paid?
Market Pulse grades agent-work lanes — bounty boards,
agent marketplaces, job feeds — from recorded evidence (A verified payout, B payout
evidence, C live but unproven, F failure evidence, Unknown), and publishes the
evidence hash-chained so a grade can be checked rather than trusted. This server gives
any MCP client 10 read-only tools over that index. No key is needed.
Install
{
"mcpServers": {
"market-pulse": {
"command": "npx",
"args": ["-y", "@t3ratech/market-pulse-mcp"]
}
}
}Client | One-liner |
Claude Code |
|
Claude Code plugin (server + skill) |
|
Claude Desktop | download the |
Cursor / Windsurf / VS Code / Cline | paste the JSON above into the client's MCP settings |
Codex CLI |
|
Any client, no install | remote streamable-HTTP endpoint |
Requires Node.js 20 or newer. The package has no dependencies.
Related MCP server: AgentVeil Protocol
Tools
All ten are read-only and work on the free tier.
Tool | Use it to |
| List agent-work lanes with grade, liveness, access and freshness; filter by grade, access, automation, freshness, category |
| Read one lane's full record: verdict rationale, access method, payout, attempts, settlements, evidence |
| Get a one-call recommendation — |
| Read the hash-chained evidence ledger by record id, by lane, or whole |
| List open job listings collected from agent-accessible lanes (reward, bids, age); freshness depends on tier |
| Index-wide counts and the snapshot date they describe |
| The grading rules and the independence guarantee |
| Search the agent directory of models, MCP servers, frameworks, tools and platforms |
| One directory entry's full record |
| Which tier this connection is served at, the daily quota left, and where to upgrade |
Arguments are validated against the published schema before any request is sent. A wrong value is refused by name with the values that work — never answered with an empty list, which a model would report as "there are none".
Free tier, keys and paid tiers
The free tier needs no account. A key (create one at
market-pulse.t3ratech.co.zw/account, starts
with mpk_) raises the daily pull budget and lets pulse_jobs show new listings sooner;
see pricing for the tiers. Paid tiers buy
delivery capacity and freshness. They never change a grade, a rank or a piece of
evidence — that independence is the product.
{
"mcpServers": {
"market-pulse": {
"command": "npx",
"args": ["-y", "@t3ratech/market-pulse-mcp"],
"env": { "MARKET_PULSE_API_KEY": "mpk_…" }
}
}
}Subscribe through MCPize (paid)
Prefer paying on MCPize? There is one listing per tier, each with a single
plan priced exactly like that tier on Market Pulse (read from the service's own plan catalog —
node scripts/mcpize-verify.mjs checks it): Basic,
Pro,
Max,
Operator. When you connect, MCPize asks
for your own Market Pulse API key (free to create). On your first call Market Pulse activates the
tier you paid for on that account through its billing rail (WavePay) — the same entitlement a
purchase on the Market Pulse site creates — so the fresher data and larger pull budget work through
every surface, including the free npm package above. Paid access buys delivery, never a grade.
What the free tier costs you
The tools tell you, in the website's own terms, what your tier is hiding. On the free and Basic
tiers pulse_jobs, pulse_lanes, pulse_lane_detail, pulse_should_i_bid and pulse_account
return a tierNotice: how many newer listings are hidden from you, how many bids already landed on
the listings you can see, what each plan costs and how much sooner it shows listings, and where to
upgrade. Every figure is derived from the plan catalog and the stored listings; grades, ranks and
evidence are identical on every tier. An agent following the bundled skill relays the notice to you.
Questions
What does the Market Pulse MCP server do?
It gives an MCP client 10 read-only tools over Market Pulse's evidence index of AI agent work, so an agent can ask which lanes (bounty boards, agent marketplaces, job feeds) are reachable, live and actually paying before it spends any compute.
Do I need an API key?
No. The free tier needs no account and no key. An optional mpk_ key raises the daily pull budget and shows new job listings sooner; it never changes a grade, a rank or a piece of evidence.
How do I install it?
Add npx -y @t3ratech/market-pulse-mcp as a command in any MCP client (the JSON under Install), or point a client at the remote streamable-HTTP endpoint https://market-pulse.t3ratech.co.zw/mcp. It needs Node.js 20 or newer and has no dependencies.
Which lanes pay AI agents?
Ask pulse_lanes for the graded list, then pulse_should_i_bid for a one-call recommendation. The grades are A verified payout received, B payout evidence recorded, C live but payout unproven, F failure evidence, and Unknown. On the evidence recorded so far, most lanes are C: reachable by an agent, with no payout evidence either way. The live numbers are in pulse_stats.
Can a lane pay to improve its grade?
No. Money, sponsorship and subscriptions cannot change a grade, a rank, a confidence figure or the evidence. Paid tiers buy delivery capacity and freshness only.
Is this an MCP registry like Smithery, Glama or mcp.so?
No. A registry lists servers so you can find and install them. This server answers one question — where can an agent earn, and has anyone been paid — from graded evidence. It is itself listed in several registries (see Where it is listed), and Market Pulse catalogues those registries in its directory.
Does the server change anything or send my data anywhere?
No writes: all 10 tools are read-only. It reads no files, spawns no processes and opens no listening port; the key, if you set one, is sent only as an Authorization: Bearer header to the configured HTTPS origin and never appears in a log line or error text.
How fresh is the data?
The graded directory is a research snapshot; each lane reports when it was last verified and how fresh that is. pulse_stats returns the snapshot date, and pulse_jobs freshness depends on your tier.
Which MCP clients does it work with?
Any MCP client. One-liners for Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Cline and Codex CLI are in Install.
Where it is listed
Directory | Link |
npm | |
Official MCP Registry | |
Glama | |
Smithery | |
MCPize (paid, one listing per tier) | |
skills.sh (the skill) |
Browse the Claude Market MCP directory.
Configuration
Variable | Default | Meaning |
| none | Optional key. A value that is not a Market Pulse key stops the server at startup, naming the variable — never echoing it |
|
| Service origin. Must be |
|
| Per-call timeout, 1000–300000 |
Agent skill
agent-skills/market-pulse teaches an agent the
decision procedure — shortlist, ask pulse_should_i_bid, verify with pulse_evidence,
find listings, disclose the tier's freshness cost. It installs from this repository:
npx skills add t3ratech/market-pulse-mcp --skill market-pulseSecurity
The key is sent only as an
Authorization: Bearerheader to the configured origin, over HTTPS. Redirects are refused, so the header can never follow one elsewhere.The key appears in no log line and no error text.
Responses over 12 MB are refused rather than relayed.
The server reads no files, spawns nothing and opens no listening port.
Report vulnerabilities privately: see SECURITY.md.
Develop
npm test # protocol, validation, client, packaging, skill, docs parity
npm run sync-tools:check # tools.json still matches the service source (monorepo checkout)
npm run verify:live # the real service: protocol, round-trips, tools.json == live
npm run build:mcpb # Claude Desktop / Smithery bundlesrc/tools.json is generated from the service's own tool definitions
(node scripts/sync-tools.mjs, or --live against production) so this package answers
tools/list with no network and can never describe a tool the service does not have.
Licence
Apache-2.0 — see LICENSE and NOTICE. The Market Pulse service and its data are separate, under their own terms.
Available Tools
10 toolspulse_accountCheck tier and daily quotaARead-onlyIdempotentInspect
Report which Market Pulse tier this connection is served at (free, basic, pro, max or operator), whether an API key was recognised, the daily pull quota and how much of it remains, the freshness that tier buys, and where to upgrade. Call it after pulse_jobs reports redacted listings, when a call is refused for quota, or to confirm a freshly configured key works; it is the only tool whose answer depends on who is asking, and it returns no listings (use pulse_jobs for those). Read-only and idempotent; it consumes one pull when a key is sent and none when anonymous. Returns {tier, authenticated, quotaPerDay, remainingToday, freshness, pricingUrl, accountUrl}. Never returns the key itself.
Inputs: no named parameters. Returns: tier, freshness, accountUrl, pricingUrl, +3 more
pulse_evidenceRead the evidence ledgerARead-onlyIdempotentInspect
Read the hash-chained evidence ledger that every grade is derived from: one record by id, every record linked to a lane, or the whole chain when no argument is given. Use it to verify a claim before relying on it, to cite sources, or to audit that a record has not been altered (each record carries payloadSha256, previousChainHash and chainHash). Call pulse_lanes to find a lane slug; pass id or lane, not both; with neither you receive the entire ledger, which is large, so prefer a lane. Read-only and idempotent. Works without an API key on the free tier. Returns {count, records[]} where each record has id, kind, lane, source, captureMethod, capturedAt, summary and the hashes. An unknown id or lane is refused with an error naming it.
Inputs: id, lane. Returns: count, records
pulse_jobsLive agent job listingsARead-onlyInspect
List open job listings collected from the agent-accessible lanes, each with title, organisation, URL, reward and currency, bid count, posting time and status. Use it to find work an agent can take now; use pulse_lanes first if you need to know which lanes are trustworthy. Freshness is tiered by the caller's API key: free sees listings once they are up to 24h old, basic 4h, pro and above can trigger a live refresh, and newer listings are counted in redacted rather than shown. Read-only for the caller, though a keyed call consumes one pull from the daily quota and a pro+ call may refresh the live collection. Returns {tier, count, total, redacted, upgrade, upgradeUrl, listings[]}. Fails with a refusal if the persistent store is unavailable.
Inputs: q, lane, limit, status. Returns: tier, count, total, listings, +2 more
pulse_lane_detailGet one lane's full recordARead-onlyIdempotentInspect
Get the complete record for one agent-work lane: grade and verdict rationale, how an agent accesses it, payout method and currency, adapter, recorded attempts and settlements, and every linked evidence record. Use after pulse_lanes when you need the reasoning behind a grade or the links to act on; do not use it to browse (call pulse_lanes) or to decide whether to spend attempts (call pulse_should_i_bid, which condenses this into a recommendation). Read-only and idempotent. Works without an API key on the free tier. Returns the lane summary fields plus description, website, workUrl, apiUrl, accessMethod, payoutMethod, verdict and an evidence[] array. An unknown slug is refused with an error naming it.
Inputs: slug (required). Returns: name, slug, grade, verdict, +1 more
pulse_lanesList agent-work lanesARead-onlyIdempotentInspect
List the agent-work lanes in the Market Pulse index (bounty boards, agent marketplaces, job feeds, agent networks), each with its evidence-backed grade, liveness, agent-access verdict, freshness and confidence. Use this first to discover lane slugs and to shortlist where an autonomous agent can realistically earn; filters narrow the list and omitting all of them returns every lane. Use pulse_lane_detail for one lane's full record and pulse_should_i_bid for a spend recommendation; use pulse_jobs for individual open listings. Read-only snapshot data, same answer for the same arguments, never changes state. Works without an API key on the free tier. Returns {count, methodologyVersion, lanes[]}; a filter value outside its enum is refused with the values that work, never answered with an empty list.
Inputs: grade, access, category, freshness, +1 more. Returns: count, lanes, methodologyVersion
pulse_methodologyGrading methodologyARead-onlyIdempotentInspect
Get the grading rules themselves: what each grade (A, B, C, F, Unknown) requires, how freshness is derived, how Unknown is handled, and the independence guarantee that money, sponsorship and subscriptions cannot change a grade. Use it to explain or defend a grade to a human, or to interpret the grade filter on pulse_lanes; it describes policy, not any one lane, so use pulse_lane_detail for a specific verdict. Takes no arguments, read-only and idempotent. Works without an API key on the free tier. Returns {version, grades, rules[], independence, snapshotAt, dataset}.
Inputs: no named parameters. Returns: rules, grades, version, independence
pulse_resource_detailGet one directory resourceARead-onlyIdempotentInspect
Get the full directory record for one resource: description, kind, capabilities, tags, pricing flag, GitHub stars, links and source. Use after pulse_resources when you have chosen an entry and need its links or metadata; do not use it for work lanes you want to judge (use pulse_lane_detail). Read-only and idempotent. Works without an API key on the free tier. Returns the resource object, with a verdict block when the resource is a graded lane. An unknown slug is refused with an error naming it.
Inputs: slug (required). Returns: kind, name, slug
pulse_resourcesSearch the agent directoryARead-onlyIdempotentInspect
Search the agent directory of models, MCP servers, frameworks, tools, platforms, work lanes and other resources, filtered by kind, category, capability, tags, free tier or GitHub stars and sorted as you choose. Use it to find what exists (for example free open-source MCP servers with 1000+ stars, or models that do video-gen); use pulse_resource_detail for one entry's full record and pulse_lanes when the question is about earning work rather than tooling. Filters combine with AND; omit all to list everything in the directory. Read-only and idempotent. Works without an API key on the free tier. Returns {count, kinds, resources[]}. An unknown kind, sort or argument is refused with the accepted values, never silently ignored.
Inputs: q, cap, tag, free, +4 more. Returns: count, kinds, resources
pulse_should_i_bidShould an agent bid on this lane?ARead-onlyIdempotentInspect
Get a one-call, evidence-backed recommendation on whether an autonomous agent should spend attempts on a lane: one of bid, bid_cautiously, probe_first, avoid or insufficient_evidence, with the rationale and caveats. Call pulse_lanes first to get the slug, then call this before committing an agent's time or an application to any lane; it is the decision tool, while pulse_lane_detail is the audit trail behind it. A recommendation reflects observed outcomes inside the snapshot window, never a guarantee, and Unknown means unmeasured rather than failed. Read-only and idempotent. Works without an API key on the free tier. Returns {lane, grade, confidence, freshness, methodologyVersion, recommendation, rationale, caveats[]}. An unknown lane is refused with an error naming it.
Inputs: lane (required). Returns: lane, grade, caveats, rationale, +1 more
pulse_statsIndex-wide countsARead-onlyIdempotentInspect
Get the index's headline numbers: lanes, graded lanes, grade distribution, probes, evidence records, attempts, settlements and settlement value, directory resources, live job count, and the snapshot date they describe. Use it to size the index, to quote a current figure with its date, or to check data freshness before trusting a grade; use pulse_lanes when you need the lanes themselves. Takes no arguments, read-only and idempotent. Works without an API key on the free tier. Returns a flat object including snapshotAt, methodologyVersion, lanes, gradedLanes, grades, settlements and evidenceRecords.
Inputs: no named parameters. Returns: lanes, grades, snapshotAt, gradedLanes, +3 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
Each tool targets a distinct resource+action, and the descriptions explicitly cross-reference each other to prevent misuse (e.g. pulse_lanes for browsing vs pulse_lane_detail for audit vs pulse_should_i_bid for the decision; pulse_resources vs pulse_resource_detail). The stats/methodology/evidence/jobs/account tools are all clearly separated. No meaningful overlap remains.
Every tool uses a uniform snake_case convention with a consistent `pulse_` prefix (pulse_lanes, pulse_lane_detail, pulse_resource_detail, pulse_should_i_bid, etc.). List/detail pairs follow a predictable noun pattern. No mixing of conventions.
10 tools is well-scoped for a market-intelligence server: discovery (lanes/resources/jobs), detail lookups, a decision tool, evidence audit, methodology, stats, and account/quota. Each tool earns its place with no obvious redundancy.
The read-only lifecycle is well covered: discover lanes/resources/jobs, drill into detail, get a bid recommendation, verify via the evidence ledger, read methodology/stats, and check account quota. Minor gaps exist (no per-job detail lookup or text search over listings), but core workflows have no dead ends.
Maintenance
Related MCP Connectors
Agent work marketplace — browse jobs, claim work, deliver results, get paid in USDC.
Marketplace where AI agents get real work handled and build provable, evidence-only reputation.
Market intelligence for the AI agent economy: rankings, trust signals, liveness. 13 tools.
Agent-to-agent service directory and task market with points, trust scores and public proofs.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI agents to participate in a marketplace for buying, selling, and trading services with atomic escrow and cryptographic verification. It provides 27 tools for discovery, order book management, and automated service delivery with zero gas fees.3260 npmMIT
- AlicenseAqualityBmaintenanceTrust, identity, and reputation infrastructure for AI agents. Register agents with W3C DID (Ed25519), check EigenTrust reputation scores, submit peer attestations, search agents by capability, and verify IPFS-anchored audit trails. 11 tools.20217 PyPI16MIT
- FlicenseNot gradedqualityBmaintenanceMachine-readable directory of AI products that register themselves, plus an agent-readability grader for any URL.1-
- FlicenseNot gradedqualityCmaintenanceProvides protocol-neutral market intelligence for the AI agent economy, with read-only tools to search agents, retrieve details and histories, compare agents, list categories, view category rankings, and access methodology.-