thinkerwave
OfficialThinkerWave
An autonomous research agent for problems without a single right answer — credit risk stress tests, pricing and churn, market entry, due diligence, drug program selection.
You give it the problem and the facts you're sure of. It researches, runs the numbers, checks its own work and tries again, round after round, keeping the best answer. You get the answer, how confident it is, and the evidence behind each claim — verified, assumption or guess — plus any charts and tables it produced.

pip install thinkerwave
export THINKERWAVE_API_KEY=tw_live_… # create one at thinkerwave.ai/app → Settings → API keys
tw run "Should we raise prices 15%, 8%, or hold? 1,200 customers at \$85/month, 3.1% monthly churn." \
--given "last 8% increase spiked churn to 4.6% for three months" --rounds 6 --watchFrom Claude Code or Cursor: pip install "thinkerwave[mcp]", then claude mcp add thinkerwave -- thinkerwave-mcp — your coding assistant can now hand ThinkerWave a hard question and read back the evidence.
From Python: see examples/.
What's in this repo: the Python SDK, the tw command line, an MCP server, the API reference and the answer/evidence format. ThinkerWave itself runs in ThinkerWave's cloud; this repository is the way in, not the engine.
Free runs are included with every account. Limits: docs/limits.md. Found a problem it got wrong? Open an issue with the run id.
Available Tools
6 toolsget_answerC
The best answer so far (or the final one), its confidence, and each finding tagged verified / assumption / guess.
Inputs: id (required). Returns: result
list_filesC
Files the run produced (charts, tables, notes).
Inputs: id (required). Returns: result
read_fileC
Read one produced file: text files as text (up to 200 KB), images as images (up to 1 MB).
Inputs: id (required), path (required).
run_statusB
Status of a run: round, best score so far, and why it stopped once finished.
Inputs: id (required). Returns: result
start_runA
Start a ThinkerWave run on a hard, open-ended problem. given: facts to keep exactly as written.
standards: what a good answer must do. pitfalls: mistakes to avoid. Returns the run id; runs take
several minutes — check back with run_status, then get_answer.
Inputs: problem (required), given, pitfalls, standards, +1 more. Returns: result
stop_runB
Stop a run. The best answer so far is kept.
Inputs: id (required). Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool maps to a distinct phase or resource: start/stop control the run lifecycle, status/answer report progress and results, and list/read handle produced artifacts. There is no meaningful overlap, and the descriptions clearly guide when to use each.
Most names follow a verb_noun snake_case pattern (start_run, stop_run, get_answer, list_files, read_file), but run_status breaks the pattern by using a noun phrase without a verb like get_. The casing and style are otherwise consistent and readable.
Six tools is well-scoped for a run-based problem-solving service. Each tool covers a distinct operation, and there are no redundant or filler tools.
The set covers the core lifecycle: start, stop, monitor status, retrieve the answer, and inspect produced files. A minor gap exists around listing existing runs or cleaning up runs/files, but agents can work around this if they retain run IDs.
Maintenance
Related MCP Connectors
Autonomous lab for the agent economy: switches parts of AI agents on and off, signs what changes.
AI agents post reproducible tests, answer questions and verify each other's results.
Free work board for autonomous agents: pick a task, deliver it, get a verifiable certificate.
Self-healing CI tests and proof of work for AI agents
Related MCP Servers
- AlicenseAqualityCmaintenanceCollective memory for AI agents. One agent solves a bug - every agent in the world gets the fix instantly.3MIT
- AlicenseNot gradedqualityFmaintenanceAI agents that hire other AI agents — and pay in SOL. Decentralized agent marketplace via Nostr + Solana.MIT
- Functional Source , Version 1.1, ALv2 Future
- AlicenseNot gradedqualityCmaintenanceAn autonomous task decomposition and execution agent that plans, executes, verifies, and commits code changes for engineering goals.MIT