ProjectBrain (projectbrain-mcp)
Local-first AST Context Compiler, Closed-World Dossier, and Surgical Anchor Patch Engine for AI Coding Agents.
ProjectBrain indexes your codebase into a typed AST & call-graph (file, fn, cls, route, mod) in ~90ms and cuts coding-agent token bills by 60%–65% while speeding up multi-file debugging by 30%–42%:
- Input Token Reduction (
-81.3%to-88.3%code tokens,-73.2%to-76.3%billed prompt tokens): Compiles module constants, smart-folded AST function slices, 1–3 hop bidirectional call chains with@L<line>caller windows, and a Closed-World Exhaustive Repo Scan into a single<800-line.projectbrain_dossier.mdread in 1read_filecall. - Output Token Reduction (
-63.2%to-77.3%edit output tokens): Replaces verbatimTarget+Replacementrepetition (edit_file) with a compact Anchor Patch DSL (@file,@sub,@replace,@insert_after,@delete) validated by Tree-sitter in<5ms. - Live ROI & Savings Ledger (
projectbrain stats): Tracks cumulative input/output tokens saved, file reads avoided, estimated USD saved, and developer wait-time saved in.projectbrain_ledger.json.
Quick Start (60 Seconds)
pip install projectbrain-mcp
projectbrain --path /path/to/repo init
Or run as a zero-install MCP server via uvx:
{
"mcpServers": {
"projectbrain": {
"command": "uvx",
"args": ["projectbrain-mcp"]
}
}
}
projectbrain init indexes the repository, updates .gitignore, writes .mcp.json (for Claude Code, Cursor, Windsurf, Antigravity), and creates .agents/rules/projectbrain.md so agents automatically use brain_dossier and brain_patch.
Verified A/B Benchmark Results (142-File Production Repo)
| Benchmark Scenario | Metric | Baseline Agent (No Brain) | ProjectBrain Agent | Improvement |
|---|---|---|---|---|
| Scenario 1: Targeted 3-Part Engineering Analysis | Code Output Ingested | 67,319 tok |
12,569 tok |
-81.3% (5.35x smaller) |
| Total Prompt Tokens Billed | 655,391 tok |
155,052 tok |
-76.3% (4.23x fewer) |
|
File Reads (read_file) |
16 |
2 |
-87.5% |
|
| Net API Cost | $0.6723 |
$0.2590 |
-61.5% (2.60x cheaper) |
|
| Scenario 2: Unseen Open-Ended Bug Report (Zero Identifiers) | Code Output Ingested | 351,658 tok (1.40M chars) |
40,982 tok (163.9k chars) |
-88.3% (8.58x smaller) |
| Total Prompt Tokens Billed | 4,561,692 tok |
1,220,579 tok |
-73.2% (-3.34M tokens) |
|
File Reads (read_file) |
84 (19 re-reads) |
6 (0 re-reads) |
-92.9% (14.0x fewer reads) |
|
| Multi-File Patch Output Tokens | 3,359 tok (edit_file) |
529 – 1,237 tok |
-63.2% to -77.3% edit output |
|
| Net API Cost | $3.2712 |
$1.1557 |
-64.7% (2.83x cheaper) |
CLI Reference
# 1. One-command repo initialization (index + .gitignore + .mcp.json + agent rules)
projectbrain --path <repo> init
# 2. Compile a single-read (<800-line) Engineering Dossier (.projectbrain_dossier.md)
projectbrain --path <repo> dossier "<question_or_bug_1>" "<question_or_bug_2>"
# 3. Apply or dry-run a Surgical Anchor Patch spec (.projectbrain_patch.diff)
projectbrain --path <repo> patch --spec "@file src/App.tsx
@sub L84 \`!profileCompleted\` -> \`isHydrated && !profileCompleted\`" [--apply]
# 4. View Live ROI & Token Savings Dashboard
projectbrain --path <repo> stats --summary
# 5. Additional granular tools
projectbrain --path <repo> slice --summary <file:symbol_or_start-end>
projectbrain --path <repo> compile "<task>" --summary
projectbrain --path <repo> explore <symbol_or_route> [target_b]
projectbrain --path <repo> impact <file1> [file2 ...]
projectbrain --path <repo> remember <symbol_or_file> "<architectural_note>"
MCP Server (projectbrain-mcp)
Exposes native MCP tools for Claude Code, Cursor, Windsurf, and Antigravity:
brain_dossier(tasks, impact_files=None, repo_path=None)— Compiles<800-line.projectbrain_dossier.mdwith module globals, folded AST slices, call chains, and Closed-World Exhaustive Repo Scan.brain_patch(spec, apply=False, repo_path=None)— Executes or dry-runs Anchor Patch DSL (@file,@sub L<n>|*|ALL|<symbol>,@replace,@insert_after,@delete) with Tree-sitter syntax verification.brain_stats(repo_path=None, reset=False)— Returns cumulative token, dollar, and time savings from.projectbrain_ledger.json.brain_slice,brain_context,brain_explore,brain_check,brain_remember.
Metadata
Release files for projectbrain-mcp 0.2.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| projectbrain_mcp-0.2.1.tar.gz | 98.2 kB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| projectbrain_mcp-0.2.1-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 181.7 kB
Release files / projectbrain_mcp-0.2.1.tar.gz
| Download URL | projectbrain_mcp-0.2.1.tar.gz |
|---|---|
| Size | 98.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
5b3f19446128151daa98c5c1163594584893b241bb6dabb468ea8d3109343129
|
|
BLAKE2b-256 checksum How to use checksums |
ac8c7952cb64805ac04e3d1a0bf3ff7cfed7a6dbd0e8f517e313b0f0ef14214a
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.5
|
Release files / projectbrain_mcp-0.2.1-py3-none-any.whl
| Download URL | projectbrain_mcp-0.2.1-py3-none-any.whl |
|---|---|
| Size | 83.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
4a1d5483976e8591a35e9416a6043c69b7c77961a8a6a0334aabbe427070eb46
|
|
BLAKE2b-256 checksum How to use checksums |
064bc9e6763b391a1e168e3018300768ae861d4fd085e902fb6267df837ed988
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.5
|
Security for projectbrain-mcp 0.2.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.